This takes advantage of the implicit default behavior to reduce the number of
attributes, which in turns reduces compilation time. I've observed -3% in
instruction count when compiling sqlite3 amalgamation with -O0
Differential Revision: https://reviews.llvm.org/D96400
The ability to specify alignment was recently added, and it's an
important property which we should ensure is set as expected by
Clang. (Especially before making further changes to Clang's code in
this area.) But, because it's on the end of the lines, the existing
tests all ignore it.
Therefore, update all the tests to also verify the expected alignment
for atomicrmw and cmpxchg. While I was in there, I also updated uses
of 'load atomic' and 'store atomic', and added the memory ordering,
where that was missing.
Signed prefix is removed and the single word spelling is
printed for the scalar types.
Tags: #clang
Differential Revision: https://reviews.llvm.org/D96161
Pipe element type spelling for arg info metadata
should follow the same behavior as normal type spelling.
We should only use the canonical type spelling in the
base type field.
This patch also removed duplication in type handling.
Tags: #clang
Differential Revision: https://reviews.llvm.org/D96151
The included test case triggered a sign assertion on the result in
`Success()`. This was caused by the APSInt created for a bitcast
having its signedness bit inverted. The second APSInt constructor
argument is `isUnsigned`, so invert the result of
`isSignedIntegerType`.
Relanding this patch after reverting. The test case had to be updated
to be insensitive to 32/64-bit extractelement indices.
Differential Revision: https://reviews.llvm.org/D95135
The included test case triggered a sign assertion on the result in
`Success()`. This was caused by the APSInt created for a bitcast
having its signedness bit inverted. The second APSInt constructor
argument is `isUnsigned`, so invert the result of
`isSignedIntegerType`.
Differential Revision: https://reviews.llvm.org/D95135
If a function doesn't contain loops and does not call non-willreturn
functions, then it is willreturn. Loops are detected by checking
for backedges in the function. We don't attempt to handle finite
loops at this point.
Differential Revision: https://reviews.llvm.org/D94633
Change "negative" into "invalid" and put "invalid" at the beginning of
the file name, following the bulk of the invalid tests in the
SemaOpenCL directory.
Use the "invalid-" prefix only for tests that contain only invalid
constructs.
Drop the "valid" suffix for CodeGen tests, as inputs in this directory
are supposed to be valid anyway.
For a default visibility external linkage definition, dso_local is set for ELF
-fno-pic/-fpie and COFF and Mach-O. Since default clang -cc1 for ELF is similar
to -fpic ("PIC Level" is not set), this nuance causes unneeded binary format differences.
To make emitted IR similar, ELF -cc1 -fpic will default to -fno-semantic-interposition,
which sets dso_local for default visibility external linkage definitions.
To make this flip smooth and enable future (dso_local as definition default),
this patch replaces (function) `define ` with `define{{.*}} `,
(variable/constant/alias) `= ` with `={{.*}} `, or inserts appropriate `{{.*}} `.
For a definition (of most linkage types), dso_local is set for ELF -fno-pic/-fpie
and COFF, but not for Mach-O. This nuance causes unneeded binary format differences.
This patch replaces (function) `define ` with `define{{.*}} `,
(variable/constant/alias) `= ` with `={{.*}} `, or inserts appropriate `{{.*}} `
if there is an explicit linkage.
* Clang will set dso_local for Mach-O, which is currently implied by TargetMachine.cpp. This will make COFF/Mach-O and executable ELF similar.
* Eventually I hope we can make dso_local the textual LLVM IR default (write explicit "dso_preemptable" when applicable) and -fpic ELF will be similar to everything else. This patch helps move toward that goal.
As mentioned in D93793, there are quite a few places where unary `IRBuilder::CreateShuffleVector(X, Mask)` can be used
instead of `IRBuilder::CreateShuffleVector(X, Undef, Mask)`.
Let's update them.
Actually, it would have been more natural if the patches were made in this order:
(1) let them use unary CreateShuffleVector first
(2) update IRBuilder::CreateShuffleVector to use poison as a placeholder value (D93793)
The order is swapped, but in terms of correctness it is still fine.
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D93923
This patch updates IRBuilder to create insertelement/shufflevector using poison as a placeholder.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D93793
Background: Call to library arithmetic functions for div is emitted by the
compiler and it set wrong “C” calling convention for calls to these functions,
whereas library functions are declared with `spir_function` calling convention.
InstCombine optimization replaces such calls with “unreachable” instruction.
It looks like clang lacks SPIRABIInfo class which should specify default
calling conventions for “system” function calls. SPIR supports only
SPIR_FUNC and SPIR_KERNEL calling convention.
Reviewers: Erich Keane, Anastasia
Differential Revision: https://reviews.llvm.org/D92721
These lit tests now requires amdgpu-registered-target since they
use clang driver and clang driver passes an LLVM option which
is available only if amdgpu target is registered.
Change-Id: I2df31967409f1627fc6d342d1ab5cc8aa17c9c0c
This will ensure that passes that add new global variables will create them
in address space 1 once the passes have been updated to no longer default
to the implicit address space zero.
This also changes AutoUpgrade.cpp to add -G1 to the DataLayout if it wasn't
already to present to ensure bitcode backwards compatibility.
Reviewed by: arsenm
Differential Revision: https://reviews.llvm.org/D84345
Noticed while fixing unused prefix warnings - there isn't actually any diff in the loop unrolled ir between old/new pass managers any more, so the broken checks were superfluous
This differentiates the Ryzen 4000/4300/4500/4700 series APUs that were
previously included in gfx909.
Differential Revision: https://reviews.llvm.org/D90419
Change-Id: Ia901a7157eb2f73ccd9f25dbacec38427312377d
[AMDGPU] Add __builtin_amdgcn_grid_size
Similar to D76772, loads the data from the dispatch pointer. Marked invariant.
Patch also updates the openmp devicertl to use this builtin.
Reviewed By: yaxunl
Differential Revision: https://reviews.llvm.org/D90251
Permitting non-standards-driven "do the best you can" constant-folding
of array bounds is permitted solely as a GNU compatibility feature. We
should not be doing it in any language mode that is attempting to be
conforming.
From https://reviews.llvm.org/D20090 it appears the intent here was to
permit `__constant int` globals to be used in array bounds, but the
change in that patch only added half of the functionality necessary to
support that in the constant evaluator. This patch adds the other half
of the functionality and turns off constant folding for array bounds in
OpenCL.
I couldn't find any spec justification for accepting the kinds of cases
that D20090 accepts, so a reference to where in the OpenCL specification
this is permitted would be useful.
Note that this change also affects the code generation in one test:
because after 'const int n = 0' we now treat 'n' as a constant
expression with value 0, it's now a null pointer, so '(local int *)n'
forms a null pointer rather than a zero pointer.
Reviewed By: Anastasia
Differential Revision: https://reviews.llvm.org/D89520
At AMD, in an internal audit of our code, we found some corner cases
where we were not quite differentiating targets enough for some old
hardware. This commit is part of fixing that by adding three new
targets:
* The "Oland" and "Hainan" variants of gfx601 are now split out into
gfx602. LLPC (in the GPUOpen driver) and other front-ends could use
that to avoid using the shaderZExport workaround on gfx602.
* One variant of gfx703 is now split out into gfx705. LLPC and other
front-ends could use that to avoid using the
shaderSpiCsRegAllocFragmentation workaround on gfx705.
* The "TongaPro" variant of gfx802 is now split out into gfx805.
TongaPro has a faster 64-bit shift than its former friends in gfx802,
and a subtarget feature could be set up for that to take advantage of
it. This commit does not make that change; it just adds the target.
V2: Add clang changes. Put TargetParser list in order.
V3: AMDGCNGPUs table in TargetParser.cpp needs to be in GPUKind order,
so fix the GPUKind order.
Differential Revision: https://reviews.llvm.org/D88916
Change-Id: Ia901a7157eb2f73ccd9f25dbacec38427312377d
- `-cl-fp32-correctly-rounded-divide-sqrt` is already handled in a
per-instruction manner by annotating the accuracy required. There's no
need to add that fn-attr. So far, there's no in-tree backend handling
that attr and that OpenCL specific option.
- In case that out-of-tree backends are broken, this change could be
reverted if those backends could not be fixed.
Differential Revision: https://reviews.llvm.org/D88424
This reverts commit 55c4ff91bd.
Issues were introduced as discussed in https://reviews.llvm.org/D88241
where this change made previous bugs in the linker and BitCodeWriter
visible.
Make the corresponding change that was made for byval in
b7141207a4. Like byval, this requires a
bulk update of the test IR tests to include the type before this can
be mandatory.
Add address space to indirect abi info and use it for kernels.
Previously, indirect arguments assumed assumed a stack passed object
in the alloca address space using byval. A stack pointer is unsuitable
for kernel arguments, which are passed in a separate, constant buffer
with a different address space.
Start using the new byref for aggregate kernel arguments. Previously
these were emitted as raw struct arguments, and turned into loads in
the backend. These will lower identically, although with byref you now
have the option of applying an explicit alignment. In the future, a
reasonable implementation would use byref for all kernel arguments
(this would be a practical problem at the moment due to losing things
like noalias on pointer arguments).
This is mostly to avoid fighting the optimizer's treatment of
aggregate load/store. SROA and instcombine both turn aggregate loads
and stores into a long sequence of element loads and stores, rather
than the optimizable memcpy I would expect in this situation. Now an
explicit memcpy will be introduced up-front which is better understood
and helps eliminate the alloca in more situations.
This skips using byref in the case where HIP kernel pointer arguments
in structs are promoted to global pointers. At minimum an additional
patch is needed to allow coercion with indirect arguments. This also
skips using it for OpenCL due to the current workaround used to
support kernels calling kernels. Distinct function bodies would need
to be generated up front instead of emitting an illegal call.
This patch introduces 2 new address spaces in OpenCL: global_device and global_host
which are a subset of a global address space, so the address space scheme will be
looking like:
```
generic->global->host
->device
->private
->local
constant
```
Justification: USM allocations may be associated with both host and device memory. We
want to give users a way to tell the compiler the allocation type of a USM pointer for
optimization purposes. (Link to the Unified Shared Memory extension:
https://github.com/intel/llvm/blob/sycl/sycl/doc/extensions/USM/cl_intel_unified_shared_memory.asciidoc)
Before this patch USM pointer could be only in opencl_global
address space, hence a device backend can't tell if a particular pointer
points to host or device memory. On FPGAs at least we can generate more
efficient hardware code if the user tells us where the pointer can point -
being able to distinguish between these types of pointers at compile time
allows us to instantiate simpler load-store units to perform memory
transactions.
Patch by Dmitry Sidorov.
Reviewed By: Anastasia
Differential Revision: https://reviews.llvm.org/D82174
We do not thread blocks with convergent calls, but this check was missing
when we decide to insert PR Phis into it (which we only do for threading).
Differential Revision: https://reviews.llvm.org/D83936
Reviewed By: nikic
Making -g[no-]column-info opt out reduces the length of a typical CC1 command line.
Additionally, in a non-debug compile, we won't see -dwarf-column-info.
OpenCL 2.0 does not allow block arguments, primarily because it is
difficult to support function pointers on the various architectures
that OpenCL targets. Clang was still accepting them.
Rename and reuse the `err_opencl_half_param` diagnostic.
Fixes PR46324.
Differential Revision: https://reviews.llvm.org/D82313
This reverts commit defd43a5b3.
with correction to solve msan report
To solve https://bugs.llvm.org/show_bug.cgi?id=46166 where the
floating point settings in PCH files aren't compatible, rewrite
FPFeatures to use a delta in the settings rather than absolute settings.
With this patch, these floating point options can be benign.
Reviewers: rjmccall
Differential Revision: https://reviews.llvm.org/D81869
This reverts commit b55d723ed6.
Reapply Modify FPFeatures to use delta not absolute settings
To solve https://bugs.llvm.org/show_bug.cgi?id=46166 where the
floating point settings in PCH files aren't compatible, rewrite
FPFeatures to use a delta in the settings rather than absolute settings.
With this patch, these floating point options can be benign.
Reviewers: rjmccall
Differential Revision: https://reviews.llvm.org/D81869
Rather than pushing inactive cleanups for the block captures at the
entry of a full expression and activating them during the creation of
the block literal, just call pushLifetimeExtendedDestroy to ensure the
cleanups are popped at the end of the scope enclosing the block
expression.
rdar://problem/63996471
Differential Revision: https://reviews.llvm.org/D81624
Canonicalize on storing FP options in LangOptions instead of
redundantly in CodeGenOptions. Incorporate -ffast-math directly
into the values of those LangOptions rather than considering it
separately when building FPOptions. Build IR attributes from
those options rather than a mix of sources.
We should really simplify the driver/cc1 interaction here and have
the driver pass down options that cc1 directly honors. That can
happen in a follow-up, though.
Patch by Michele Scandale!
https://reviews.llvm.org/D80315
Sometimes you want to disable a FileCheck directive without removing
it entirely, or you want to write comments that mention a directive by
name. The `COM:` directive makes it easy to do this. For example,
you might have:
```
; X32: pinsrd_1:
; X32: pinsrd $1, 4(%esp), %xmm0
; COM: FIXME: X64 isn't working correctly yet for this part of codegen, but
; COM: X64 will have something similar to X32:
; COM:
; COM: X64: pinsrd_1:
; COM: X64: pinsrd $1, %edi, %xmm0
```
Without this patch, you need to use some combination of rewording and
directive syntax mangling to prevent FileCheck from recognizing the
commented occurrences of `X32:` and `X64:` above as directives.
Moreover, FileCheck diagnostics have been proposed that might complain
about the occurrences of `X64` that don't have the trailing `:`
because they look like directive typos:
<http://lists.llvm.org/pipermail/llvm-dev/2020-April/140610.html>
I think dodging all these problems can prove tedious for test authors,
and directive syntax mangling already makes the purpose of existing
test code unclear. `COM:` can avoid all these problems.
This patch also updates the small set of existing tests that define
`COM` as a check prefix:
- clang/test/CodeGen/default-address-space.c
- clang/test/CodeGenOpenCL/addr-space-struct-arg.cl
- clang/test/Driver/hip-device-libs.hip
- llvm/test/Assembler/drop-debug-info-nonzero-alloca.ll
I think lit should support `COM:` as well. Perhaps `clang -verify`
should too.
Reviewed By: jhenderson, thopre
Differential Revision: https://reviews.llvm.org/D79276
Sometimes you want to disable a FileCheck directive without removing
it entirely, or you want to write comments that mention a directive by
name. The `COM:` directive makes it easy to do this. For example,
you might have:
```
; X32: pinsrd_1:
; X32: pinsrd $1, 4(%esp), %xmm0
; COM: FIXME: X64 isn't working correctly yet for this part of codegen, but
; COM: X64 will have something similar to X32:
; COM:
; COM: X64: pinsrd_1:
; COM: X64: pinsrd $1, %edi, %xmm0
```
Without this patch, you need to use some combination of rewording and
directive syntax mangling to prevent FileCheck from recognizing the
commented occurrences of `X32:` and `X64:` above as directives.
Moreover, FileCheck diagnostics have been proposed that might complain
about the occurrences of `X64` that don't have the trailing `:`
because they look like directive typos:
<http://lists.llvm.org/pipermail/llvm-dev/2020-April/140610.html>
I think dodging all these problems can prove tedious for test authors,
and directive syntax mangling already makes the purpose of existing
test code unclear. `COM:` can avoid all these problems.
This patch also updates the small set of existing tests that define
`COM` as a check prefix:
- clang/test/CodeGen/default-address-space.c
- clang/test/CodeGenOpenCL/addr-space-struct-arg.cl
- clang/test/Driver/hip-device-libs.hip
- llvm/test/Assembler/drop-debug-info-nonzero-alloca.ll
I think lit should support `COM:` as well. Perhaps `clang -verify`
should too.
Reviewed By: jhenderson, thopre
Differential Revision: https://reviews.llvm.org/D79276
test cases
Add support for #pragma float_control
Reviewers: rjmccall, erichkeane, sepavloff
Differential Revision: https://reviews.llvm.org/D72841
This reverts commit 85dc033cac, and makes
corrections to the test cases that failed on buildbots.
Currently this asserts on anything other than errors. In one
workaround scenario, AMDGPU emits DiagnosticInfoUnsupported as a
warning for functions that can't be correctly codegened, but should
never be executed.
This reverts commit 61ba1481e2.
I'm reverting this because it breaks the lldb build with
incomplete switch coverage warnings. I would fix it forward,
but am not familiar enough with lldb to determine the correct
fix.
lldb/source/Plugins/TypeSystem/Clang/TypeSystemClang.cpp:3958:11: error: enumeration values 'DependentExtInt' and 'ExtInt' not handled in switch [-Werror,-Wswitch]
switch (qual_type->getTypeClass()) {
^
lldb/source/Plugins/TypeSystem/Clang/TypeSystemClang.cpp:4633:11: error: enumeration values 'DependentExtInt' and 'ExtInt' not handled in switch [-Werror,-Wswitch]
switch (qual_type->getTypeClass()) {
^
lldb/source/Plugins/TypeSystem/Clang/TypeSystemClang.cpp:4889:11: error: enumeration values 'DependentExtInt' and 'ExtInt' not handled in switch [-Werror,-Wswitch]
switch (qual_type->getTypeClass()) {
Introduction/Motivation:
LLVM-IR supports integers of non-power-of-2 bitwidth, in the iN syntax.
Integers of non-power-of-two aren't particularly interesting or useful
on most hardware, so much so that no language in Clang has been
motivated to expose it before.
However, in the case of FPGA hardware normal integer types where the
full bitwidth isn't used, is extremely wasteful and has severe
performance/space concerns. Because of this, Intel has introduced this
functionality in the High Level Synthesis compiler[0]
under the name "Arbitrary Precision Integer" (ap_int for short). This
has been extremely useful and effective for our users, permitting them
to optimize their storage and operation space on an architecture where
both can be extremely expensive.
We are proposing upstreaming a more palatable version of this to the
community, in the form of this proposal and accompanying patch. We are
proposing the syntax _ExtInt(N). We intend to propose this to the WG14
committee[1], and the underscore-capital seems like the active direction
for a WG14 paper's acceptance. An alternative that Richard Smith
suggested on the initial review was __int(N), however we believe that
is much less acceptable by WG14. We considered _Int, however _Int is
used as an identifier in libstdc++ and there is no good way to fall
back to an identifier (since _Int(5) is indistinguishable from an
unnamed initializer of a template type named _Int).
[0]https://www.intel.com/content/www/us/en/software/programmable/quartus-prime/hls-compiler.html)
[1]http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2472.pdf
Differential Revision: https://reviews.llvm.org/D73967
Currently the library is separately linked, but this isn't correct to
implement fast math flags correctly. Each module should get the
version of the library appropriate for its combination of fast math
and related flags, with the attributes propagated into its functions
and internalized.
HIP already maintains the list of libraries, but this is not used for
OpenCL. Unfortunately, HIP uses a separate --hip-device-lib argument,
despite both languages using the same bitcode library. Eventually
these two searches need to be merged.
An additional problem is there are 3 different locations the libraries
are installed, depending on which build is used. This also needs to be
consolidated (or at least the search logic needs to deal with this
unnecessary complexity).
The main purpose of introducing these builtins is to add a range
metadata [1, 1025) on the work group size loaded from dispatch
ptr, which cannot be done by source code.
Differential Revision: https://reviews.llvm.org/D76772
SPIRV2.0 Spec only specifies Linux mangling, however our downstream has
use for a Windows mangling for these types.
Unfortunately, the SPIRV
spec specifies a single mangling for all pipe types, despite clang
allowing overloading on these types. Because of this, this patch
chooses to mangle the read/writability and element type for the windows
mangling.
The windows manglings in the test all demangle according to demangler:
"void __cdecl test1(struct __clang::ocl_pipe<int,1>)
"void __cdecl test2(struct __clang::ocl_pipe<float,0>)
"void __cdecl test2(struct __clang::ocl_pipe<int,1>)
"void __cdecl test3(struct __clang::ocl_pipe<int const,1>)
"void __cdecl test4(struct __clang::ocl_pipe<union
__clang::__vector<unsigned char,3>,1>)
"void __cdecl test5(struct __clang::ocl_pipe<union
__clang::__vector<int,4>,1>)
"void __cdecl test_reserved_read_pipe(struct __clang::_ASCLglobal<struct
Person > * __ptr64,struct __clang::ocl_pipe<struct Person,1>)
Differential Revision: https://reviews.llvm.org/D75685
After a first attempt to fix the test-suite failures, my first recommit
caused the same failures again. I had updated CMakeList.txt files of
tests that needed -fcommon, but it turns out that there are also
Makefiles which are used by some bots, so I've updated these Makefiles
now too.
See the original commit message for more details on this change:
0a9fc9233e
This includes fixes for:
- test-suite: some benchmarks need to be compiled with -fcommon, see D75557.
- compiler-rt: one test needed -fcommon, and another a change, see D75520.
This reverts commit 737394c490.
The fp-model test was failing on platforms that enable denormal flushing
based on -ffast-math. This needs to reset to IEEE, not the default in
these cases.
Change-Id: Ibbad32f66d0d0b89b9c1173a3a96fb1a570ddd89
The IR hasn't switched the default yet, so explicitly add the ieee
attributes.
I'm still not really sure how the target default denormal mode should
interact with -fno-unsafe-math-optimizations. The target may have
selected the default mode to be non-IEEE based on the flags or based
on its true behavior, but we don't know which is the case. Since the
only users of a non-IEEE mode without a flag still support IEEE mode,
just reset to IEEE.
This reverts commit 0a9fc9233e.
Going to look at the asan failures.
I find the failures in the test suite weird, because they look
like compile time test and I don't understand how that can be
failing, but will have a brief look at that too.
This makes -fno-common the default for all targets because this has performance
and code-size benefits and is more language conforming for C code.
Additionally, GCC10 also defaults to -fno-common and so we get consistent
behaviour with GCC.
With this change, C code that uses tentative definitions as definitions of a
variable in multiple translation units will trigger multiple-definition linker
errors. Generally, this occurs when the use of the extern keyword is neglected
in the declaration of a variable in a header file. In some cases, no specific
translation unit provides a definition of the variable. The previous behavior
can be restored by specifying -fcommon.
As GCC has switched already, we benefit from applications already being ported
and existing documentation how to do this. For example:
- https://gcc.gnu.org/gcc-10/porting_to.html
- https://wiki.gentoo.org/wiki/Gcc_10_porting_notes/fno_common
Differential revision: https://reviews.llvm.org/D75056
norecurse function attr indicates the function is not called recursively
directly or indirectly.
Add norecurse to OpenCL functions, SYCL functions in device compilation
and CUDA/HIP kernels.
Although there is LLVM pass adding norecurse to functions, it only works
for whole-program compilation. Also FE adding norecurse can make that
pass run faster since functions with norecurse do not need to be checked
again.
Differential Revision: https://reviews.llvm.org/D73651
This CL adds clang declarations of built-in functions for AMDGPU MFMA intrinsics and instructions.
OpenCL tests for new built-ins are included.
Differential Revision: https://reviews.llvm.org/D72723
Currently there are 4 different mechanisms for controlling denormal
flushing behavior, and about as many equivalent frontend controls.
- AMDGPU uses the fp32-denormals and fp64-f16-denormals subtarget features
- NVPTX uses the nvptx-f32ftz attribute
- ARM directly uses the denormal-fp-math attribute
- Other targets indirectly use denormal-fp-math in one DAGCombine
- cl-denorms-are-zero has a corresponding denorms-are-zero attribute
AMDGPU wants a distinct control for f32 flushing from f16/f64, and as
far as I can tell the same is true for NVPTX (based on the attribute
name).
Work on consolidating these into the denormal-fp-math attribute, and a
new type specific denormal-fp-math-f32 variant. Only ARM seems to
support the two different flush modes, so this is overkill for the
other use cases. Ideally we would error on the unsupported
positive-zero mode on other targets from somewhere.
Move the logic for selecting the flush mode into the compiler driver,
instead of handling it in cc1. denormal-fp-math/denormal-fp-math-f32
are now both cc1 flags, but denormal-fp-math-f32 is not yet exposed as
a user flag.
-cl-denorms-are-zero, -fcuda-flush-denormals-to-zero and
-fno-cuda-flush-denormals-to-zero will be mapped to
-fp-denormal-math-f32=ieee or preserve-sign rather than the old
attributes.
Stop emitting the denorms-are-zero attribute for the OpenCL flag. It
has no in-tree users. The meaning would also be target dependent, such
as the AMDGPU choice to treat this as only meaning allow flushing of
f32 and not f16 or f64. The naming is also potentially confusing,
since DAZ in other contexts refers to instructions implicitly treating
input denormals as zero, not necessarily flushing output denormals to
zero.
This also does not attempt to change the behavior for the current
attribute. The LangRef now states that the default is ieee behavior,
but this is inaccurate for the current implementation. The clang
handling is slightly hacky to avoid touching the existing
denormal-fp-math uses. Fixing this will be left for a future patch.
AMDGPU is still using the subtarget feature to control the denormal
mode, but the new attribute are now emitted. A future change will
switch this and remove the subtarget features.
Commit 9a8d477a0e ("[OpenCL] Add builtin function attribute
handling", 2019-11-05) stopped Clang from mangling single-overload
builtins, which is incorrect.
Add handling for the "pure", "const" and "convergent" function
attributes for OpenCL builtin functions.
Patch by Pierre Gondois and Sven van Haastregt.
Differential Revision: https://reviews.llvm.org/D64319
The implementation of the OpenCL builtin currently library uses 2
different hacks to get to the corresponding IR intrinsics from the
source. This will allow removal of those.
This is the set that is currently used (minus a few vector ones).
llvm-svn: 367973
Stanislav Mekhanoshin