This patch fixes an issue where Microsoft-specific layout attributes,
such as __declspec(empty_bases), were ignored during CUDA/HIP device
compilation on a Windows host. This caused a critical memory layout
mismatch between host and device objects, breaking libraries that rely
on these attributes for ABI compatibility.
The fix introduces a centralized hasMicrosoftRecordLayout() check within
the TargetInfo class. This check is aware of the auxiliary (host) target
and is set during TargetInfo::adjust if the host uses a Microsoft ABI.
The empty_bases, layout_version, and msvc::no_unique_address attributes
now use this centralized flag, ensuring device code respects them and
maintains layout consistency with the host.
Fixes: https://github.com/llvm/llvm-project/issues/146047
We have multiple different attributes in clang representing device
kernels for specific targets/languages. Refactor them into one attribute
with different spellings to make it more easily scalable for new
languages/targets.
---------
Signed-off-by: Sarnie, Nick <nick.sarnie@intel.com>
This both reapplies #118734, the initial attempt at this, and updates it
significantly.
First, it uses the newly added `StringTable` abstraction for string
tables, and simplifies the construction to build the string table and
info arrays separately. This should reduce any `constexpr` compile time
memory or CPU cost of the original PR while significantly improving the
APIs throughout.
It also restructures the builtins to support sharding across several
independent tables. This accomplishes two improvements from the
original PR:
1) It improves the APIs used significantly.
2) When builtins are defined from different sources (like SVE vs MVE in
AArch64), this allows each of them to build their own string table
independently rather than having to merge the string tables and info
structures.
3) It allows each shard to factor out a common prefix, often cutting the
size of the strings needed for the builtins by a factor two.
The second point is important both to allow different mechanisms of
construction (for example a `.def` file and a tablegen'ed `.inc` file,
or different tablegen'ed `.inc files), it also simply reduces the sizes
of these tables which is valuable given how large they are in some
cases. The third builds on that size reduction.
Initially, we use this new sharding rather than merging tables in
AArch64, LoongArch, RISCV, and X86. Mostly this helps ensure the system
works, as without further changes these still push scaling limits.
Subsequent commits will more deeply leverage the new structure,
including using the prefix capabilities which cannot be easily factored
out here and requires deep changes to the targets.
Reverts llvm/llvm-project#118734
There are currently some specific versions of MSVC that are miscompiling
this code (we think). We don't know why as all the other build bots and
at least some folks' local Windows builds work fine.
This is a candidate revert to help the relevant folks catch their
builders up and have time to debug the issue. However, the expectation
is to roll forward at some point with a workaround if at all possible.
The Clang binary (and any binary linking Clang as a library), when built
using PIE, ends up with a pretty shocking number of dynamic relocations
to apply to the executable image: roughly 400k.
Each of these takes up binary space in the executable, and perhaps most
interestingly takes start-up time to apply the relocations.
The largest pattern I identified were the strings used to describe
target builtins. The addresses of these string literals were stored into
huge arrays, each one requiring a dynamic relocation. The way to avoid
this is to design the target builtins to use a single large table of
strings and offsets within the table for the individual strings. This
switches the builtin management to such a scheme.
This saves over 100k dynamic relocations by my measurement, an over 25%
reduction. Just looking at byte size improvements, using the `bloaty`
tool to compare a newly built `clang` binary to an old one:
```
FILE SIZE VM SIZE
-------------- --------------
+1.4% +653Ki +1.4% +653Ki .rodata
+0.0% +960 +0.0% +960 .text
+0.0% +197 +0.0% +197 .dynstr
+0.0% +184 +0.0% +184 .eh_frame
+0.0% +96 +0.0% +96 .dynsym
+0.0% +40 +0.0% +40 .eh_frame_hdr
+114% +32 [ = ] 0 [Unmapped]
+0.0% +20 +0.0% +20 .gnu.hash
+0.0% +8 +0.0% +8 .gnu.version
+0.9% +7 +0.9% +7 [LOAD #2 [R]]
[ = ] 0 -75.4% -3.00Ki .relro_padding
-16.1% -802Ki -16.1% -802Ki .data.rel.ro
-27.3% -2.52Mi -27.3% -2.52Mi .rela.dyn
-1.6% -2.66Mi -1.6% -2.66Mi TOTAL
```
We get a 16% reduction in the `.data.rel.ro` section, and nearly 30%
reduction in `.rela.dyn` where those reloctaions are stored.
This is also visible in my benchmarking of binary start-up overhead at
least:
```
Benchmark 1: ./old_clang --version
Time (mean ± σ): 17.6 ms ± 1.5 ms [User: 4.1 ms, System: 13.3 ms]
Range (min … max): 14.2 ms … 22.8 ms 162 runs
Benchmark 2: ./new_clang --version
Time (mean ± σ): 15.5 ms ± 1.4 ms [User: 3.6 ms, System: 11.8 ms]
Range (min … max): 12.4 ms … 20.3 ms 216 runs
Summary
'./new_clang --version' ran
1.13 ± 0.14 times faster than './old_clang --version'
```
We get about 2ms faster `--version` runs. While there is a lot of noise
in binary execution time, this delta is pretty consistent, and
represents over 10% improvement. This is particularly interesting to me
because for very short source files, repeatedly starting the `clang`
binary is actually the dominant cost. For example, `configure` scripts
running against the `clang` compiler are slow in large part because of
binary start up time, not the time to process the actual inputs to the
compiler.
----
This PR implements the string tables using `constexpr` code and the
existing macro system. I understand that the builtins are moving towards
a TableGen model, and if complete that would provide more options for
modeling this. Unfortunately, that migration isn't complete, and even
the parts that are migrated still rely on the ability to break out of
the TableGen model and directly expand an X-macro style `BUILTIN(...)`
textually. I looked at trying to complete the move to TableGen, but it
would both require the difficult migration of the remaining targets, and
solving some tricky problems with how to move away from any macro-based
expansion.
I was also able to find a reasonably clean and effective way of doing
this with the existing macros and some `constexpr` code that I think is
clean enough to be a pretty good intermediate state, and maybe give a
good target for the eventual TableGen solution. I was also able to
factor the macros into set of consistent patterns that avoids a
significant regression in overall boilerplate.
Summary:
Previous changes relaxed the address space rules based on what the
target says about them. This accidentally included the AS(2) region as
convertible to generic. Simply check for AS(2) and reject it.
Summary:
Address spaces are used in several embedded and GPU targets to describe
accesses to different types of memory. Currently we use the address
space enumerations to control which address spaces are considered
supersets of eachother, however this is also a target level property as
described by the C standard's passing mentions. This patch allows the
address space checks to use the target information to decide if a
pointer conversion is legal. For AMDGPU and NVPTX, all supported address
spaces can be converted to the default address space.
More semantic checks can be added on top of this, for now I'm mainly
looking to get more standard semantics working for C/C++. Right now the
address space conversions must all be done explicitly in C/C++ unlike
the offloading languages which define their own custom address spaces
that just map to the same target specific ones anyway. The main question
is if this behavior is a function of the target or the language.
Summary:
The cache line size on AMDGPU varies between 64 and 128 (The lowest L2
cache also goes to 256 on some architectures.) This macro is intended to
present a size that will not cause destructive interference, so we
choose the larger of those values.
- Address space cast of nullptr in local_space into a generic_space for
the CUDA backend. The reason for this cast was having invalid local
memory base address for the associated variable.
- In the context of AMD GPU, assigns a NULL value as ~0 for the address
spaces of sycl_local and sycl_private to match the ones for opencl_local
and opencl_private.
Types of AMDGPU address space were defined not only in Clang-specific class
but also in LLVM header.
If we unify the AMD GPU address space enumeration, then we can reuse it in
Clang, Flang and LLVM.
HIP texture/image support is optional as some devices
do not have image instructions. A macro __HIP_NO_IMAGE_SUPPORT
is defined for device not supporting images (d0448aa4c4/docs/reference/kernel_language.md (L426) )
Currently the macro is defined by HIP header based on predefined macros
for GPU, e.g __gfx*__ , which is error prone. This patch let clang
emit the predefined macro.
Reviewed by: Matt Arsenault, Artem Belevich
Differential Revision: https://reviews.llvm.org/D151349
Pursuant to discussions at
https://discourse.llvm.org/t/rfc-c-23-p1467r9-extended-floating-point-types-and-standard-names/70033/22,
this commit enhances the handling of the __bf16 type in Clang.
- Firstly, it upgrades __bf16 from a storage-only type to an arithmetic
type.
- Secondly, it changes the mangling of __bf16 to DF16b on all
architectures except ARM. This change has been made in
accordance with the finalization of the mangling for the
std::bfloat16_t type, as discussed at
https://github.com/itanium-cxx-abi/cxx-abi/pull/147.
- Finally, this commit extends the existing excess precision support to
the __bf16 type. This applies to hardware architectures that do not
natively support bfloat16 arithmetic.
Appropriate tests have been added to verify the effects of these
changes and ensure no regressions in other areas of the compiler.
Reviewed By: rjmccall, pengfei, zahiraam
Differential Revision: https://reviews.llvm.org/D150913
Predefine __AMDGCN_CUMODE__ as 1 or 0 when compilation assumes CU or WGP modes.
If WGP mode is not supported, ignore -mno-cumode and emit a warning.
This is needed for implementing device functions like __smid
(312dff7b79/include/hip/amd_detail/amd_device_functions.h (L957))
Reviewed by: Matt Arsenault, Artem Belevich, Brian Sumner
Differential Revision: https://reviews.llvm.org/D145343
Change the return type of `getClobbers` function from `const char*`
to `std::string_view`. Update the function usages in CodeGen module.
The reasoning of these changes is to remove unsafe `const char*`
strings and prevent unnecessary allocations for constructing the
`std::string` in usages of `getClobbers()` function.
Differential Revision: https://reviews.llvm.org/D148799
This avoids recomputing string length that is already known at compile time.
It has a slight impact on preprocessing / compile time, see
https://llvm-compile-time-tracker.com/compare.php?from=3f36d2d579d8b0e8824d9dd99bfa79f456858f88&to=e49640c507ddc6615b5e503144301c8e41f8f434&stat=instructions:u
This a recommit of e953ae5bbc313fd0cc980ce021d487e5b5199ea4 and the subsequent fixes caa713559bd38f337d7d35de35686775e8fb5175 and 06b90e2e9c991e211fecc97948e533320a825470.
The above patchset caused some version of GCC to take eons to compile clang/lib/Basic/Targets/AArch64.cpp, as spotted in aa171833ab0017d9732e82b8682c9848ab25ff9e.
The fix is to make BuiltinInfo tables a compilation unit static variable, instead of a private static variable.
Differential Revision: https://reviews.llvm.org/D139881
- [Clang] Declare AMDGPU target as supporting BF16 for storage-only purposes on amdgcn
- Add Sema & CodeGen tests cases.
- Also add cases that D138651 would have covered as this patch replaces it.
- [AMDGPU] Add BF16 storage-only support
- Support legalization/dealing with bf16 operations in DAGIsel.
- bf16 as a type remains illegal and is represented as i16 for storage purposes.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D139398
This patch mechanically replaces None with std::nullopt where the
compiler would warn if None were deprecated. The intent is to reduce
the amount of manual work required in migrating from Optional to
std::optional.
This is part of an effort to migrate from llvm::Optional to
std::optional:
https://discourse.llvm.org/t/deprecating-llvm-optional-x-hasvalue-getvalue-getvalueor/63716
Mixing LLVM and Clang address spaces can result in subtle bugs, and there
is no need for this hook to use the LLVM IR level address spaces.
Most of this change is just replacing zero with LangAS::Default,
but it also allows us to remove a few calls to getTargetAddressSpace().
This also removes a stale comment+workaround in
CGDebugInfo::CreatePointerLikeType(): ASTContext::getTypeSize() does
return the expected size for ReferenceType (and handles address spaces).
Differential Revision: https://reviews.llvm.org/D138295
Allows emitting define amdgpu_kernel void @func() IR from C or C++.
This replaces the current workflow which is to write a stub in opencl that
calls an external C function implemented in C++ combined through llvm-link.
Calling the resulting function still requires a manual implementation of the
ABI from the host side. The primary application is for more rapid debugging
of the amdgpu backend by permuting a C or C++ test file instead of manually
updating an IR file.
Implementation closely follows D54425. Non-amd reviewers from there.
Reviewed By: yaxunl
Differential Revision: https://reviews.llvm.org/D125970
WG14 adopted the _ExtInt feature from Clang for C23, but renamed the
type to be _BitInt. This patch does the vast majority of the work to
rename _ExtInt to _BitInt, which accounts for most of its size. The new
type is exposed in older C modes and all C++ modes as a conforming
extension. However, there are functional changes worth calling out:
* Deprecates _ExtInt with a fix-it to help users migrate to _BitInt.
* Updates the mangling for the type.
* Updates the documentation and adds a release note to warn users what
is going on.
* Adds new diagnostics for use of _BitInt to call out when it's used as
a Clang extension or as a pre-C23 compatibility concern.
* Adds new tests for the new diagnostic behaviors.
I want to call out the ABI break specifically. We do not believe that
this break will cause a significant imposition for early adopters of
the feature, and so this is being done as a full break. If it turns out
there are critical uses where recompilation is not an option for some
reason, we can consider using ABI tags to ease the transition.
Lets wavefront size be 32 for amdgpu openmp, as well as 64.
Fixes up as little as possible to pass that through the libraries. This change
is end to end, as opposed to updating clang/devicertl/plugin separately. It can
be broken up for review/commit if preferred. Posting as-is so that others with
a gfx10 can try it out. It works roughly as well as gfx9 for me, but there are
probably bugs remaining as well as the todo: for letting grid values vary more.
Reviewed By: ronlieb
Differential Revision: https://reviews.llvm.org/D108708
Remove redundant fields and replace pointer with virtual function
Of fourteen fields, three are dead and four can be computed from the
remainder. This leaves a couple of currently dead fields in place as
they are expected to be used from the deviceRTL shortly. Two of the
fields that can be computed are only used from codegen and require a
log2() implementation so are inlined into codegen instead.
This change leaves the new methods in the same location in the struct
as the previous fields for convenience at review.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D108380
Remove redundant fields and replace pointer with virtual function
Of fourteen fields, three are dead and four can be computed from the
remainder. This leaves a couple of currently dead fields in place as
they are expected to be used from the deviceRTL shortly. Two of the
fields that can be computed are only used from codegen and require a
log2() implementation so are inlined into codegen instead.
This change leaves the new methods in the same location in the struct
as the previous fields for convenience at review.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D108380
This patch allows target specific addr space in target builtins for HIP. It inserts implicit addr
space cast for non-generic pointer to generic pointer in general, and inserts implicit addr
space cast for generic to non-generic for target builtin arguments only.
It is NFC for non-HIP languages.
Differential Revision: https://reviews.llvm.org/D102405
This patch allows target specific addr space in target builtins for HIP. It inserts implicit addr
space cast for non-generic pointer to generic pointer in general, and inserts implicit addr
space cast for generic to non-generic for target builtin arguments only.
It is NFC for non-HIP languages.
Differential Revision: https://reviews.llvm.org/D102405
This feature requires support of __opencl_c_images, so diagnostics for that is provided as well.
Also, ensure that cl_khr_3d_image_writes feature macro is set to the same value.
Reviewed By: Anastasia
Differential Revision: https://reviews.llvm.org/D106260
Allow use of bit-fields as a clang extension
in OpenCL. The extension can be enabled using
pragma directives.
This fixes PR45339!
Differential Revision: https://reviews.llvm.org/D101843
There already exists cl_khr_fp64 extension. So OpenCL C 3.0
and higher should use the feature, earlier versions still
use the extension. OpenCL C 3.0 API spec states that extension
will be not described in the option string if corresponding
optional functionality is not supported (see 4.2. Querying Devices).
Due to that fact the usage of features for OpenCL C 3.0 must
be as follows:
```
$ clang -Xclang -cl-ext=+cl_khr_fp64,+__opencl_c_fp64 ...
$ clang -Xclang -cl-ext=-cl_khr_fp64,-__opencl_c_fp64 ...
```
e.g. the feature and the equivalent extension (if exists)
must be set to the same values
Reviewed By: Anastasia
Differential Revision: https://reviews.llvm.org/D96524
Added __cl_clang_non_portable_kernel_param_types extension that
allows using non-portable types as kernel parameters. This allows
bypassing the portability guarantees from the restrictions specified
in C++ for OpenCL v1.0 s2.4.
Currently this only disables the restrictions related to the data
layout. The programmer should ensure the compiler generates the same
layout for host and device or otherwise the argument should only be
accessed on the device side. This extension could be extended to other
case (e.g. permitting size_t) if desired in the future.
Patch by olestrohm (Ole Strohm)!
https://reviews.llvm.org/D101168
