Fast FMAF is not a sufficient condition to enable denormals.
Before VI, enabling denormals caused F32 instructions to
run at F64 speeds.
llvm-svn: 339278
Always emit alloca in entry block for enqueue_kernel builtin.
Ensures the statically sized alloca is not converted to DYNAMIC_STACKALLOC
later because it is not in the entry block.
llvm-svn: 339150
Ensures the statically sized alloca is not converted to DYNAMIC_STACKALLOC
later because it is not in the entry block.
Differential Revision: https://reviews.llvm.org/D50104
llvm-svn: 338899
The way address space declarations for builtins currently work
is nearly useless. The code assumes the address spaces used for
builtins is a confusingly named "target address space" from user
code using __attribute__((address_space(N))) that matches
the builtin declaration. There's no way to use this to declare
a builtin that returns a language specific address space.
The terminology used is highly cofusing since it has nothing
to do with the the address space selected by the target to use
for a language address space.
This feature is essentially unused as-is. AMDGPU and NVPTX
are the only in-tree targets attempting to use this. The AMDGPU
builtins certainly do not behave as intended (i.e. all of the
builtins returning pointers can never compile because the numbered
address space never matches the expected named address space).
The NVPTX builtins are missing tests for some, and the others
seem to rely on an implicit addrspacecast.
Change the used address space for builtins based on a target
hook to allow using a language address space for a builtin.
This allows the same builtin declaration to be used for multiple
languages with similarly purposed address spaces (e.g. the same
AMDGPU builtin can be used in OpenCL and CUDA even though the
constant address spaces are arbitarily different).
This breaks the possibility of using arbitrary numbered
address spaces alongside the named address spaces for builtins.
If this is an issue we probably need to introduce another builtin
declaration character to distinguish language address spaces from
so-called "target address spaces".
llvm-svn: 338707
OpenCL block literal structs have different fields which are now correctly
identified in the debug info.
Differential Revision: https://reviews.llvm.org/D49930
llvm-svn: 338299
Summary: Automatic variable initialization was generating default-aligned stores (which are deprecated) instead of using the known alignment from the alloca. Further, they didn't specify inbounds.
Subscribers: dexonsmith, cfe-commits
Differential Revision: https://reviews.llvm.org/D49209
llvm-svn: 337041
1. added restrictions to memory scope, order and volatile parameters
2. added custom processing for these builtins - currently is not used code,
needed to switch off GCCBuiltin link to the builtins (ongoing change to llvm
tree)
3. builtins renamed as requested
Differential Revision: https://reviews.llvm.org/D43281
llvm-svn: 332848
There shouldn't be any tests that run the entire optimizer here,
but the last test in this file is definitely going to break with
a change in LLVM IR canonicalization. Change that part to check
the unoptimized IR because that's the real intent of this file.
llvm-svn: 332473
Added string literal helper function to obtain the type
attributed by a constant address space.
Also fixed predefind __func__ expr to use the helper
to constract the string literal correctly.
Differential Revision: https://reviews.llvm.org/D46049
llvm-svn: 331877
Half-type mangling is accomplished following the method introduced by Erich
Keane for mangling _Float16. Updated the half.cl LIT test to cover this
particular case.
Patch By: vbridgers
Differential Revision: https://reviews.llvm.org/D46131
llvm-svn: 331263
SPIR-V encodes the read_only and write_only access qualifiers of pipes,
so separate LLVM IR types are required to target SPIR-V. Other backends
may also find this useful.
These new types are `opencl.pipe_ro_t` and `opencl.pipe_wo_t`, which
replace `opencl.pipe_t`.
This replaces __get_pipe_num_packets(...) and __get_pipe_max_packets(...)
which took a read_only pipe with separate versions for read_only and
write_only pipes, namely:
* __get_pipe_num_packets_ro(...)
* __get_pipe_num_packets_wo(...)
* __get_pipe_max_packets_ro(...)
* __get_pipe_max_packets_wo(...)
These separate versions exist to avoid needing a bitcast to one of the
two qualified pipe types.
Patch by Stuart Brady.
Differential Revision: https://reviews.llvm.org/D46015
llvm-svn: 331026
Found via codespell -q 3 -I ../clang-whitelist.txt
Where whitelist consists of:
archtype
cas
classs
checkk
compres
definit
frome
iff
inteval
ith
lod
methode
nd
optin
ot
pres
statics
te
thru
Patch by luzpaz! (This is a subset of D44188 that applies cleanly with a few
files that have dubious fixes reverted.)
Differential revision: https://reviews.llvm.org/D44188
llvm-svn: 329399
Need to override convertConstraint to recognise amdgpu specific register names.
Differential Revision: https://reviews.llvm.org/D44533
llvm-svn: 328359
Add two additional implicit arguments for OpenCL for the AMDGPU target using the AMDHSA runtime to support device enqueue.
Differential Revision: https://reviews.llvm.org/D44696
llvm-svn: 328350
- Remove use of the opencl and amdopencl environment member of the target triple for the AMDGPU target.
- Use a function attribute to communicate to the AMDGPU backend.
Differential Revision: https://reviews.llvm.org/D43735
llvm-svn: 328347
The indirect function argument is in alloca address space in LLVM IR. However,
during Clang codegen for C++, the address space of indirect function argument
should match its address space in the source code, i.e., default addr space, even
for indirect argument. This is because destructor of the indirect argument may
be called in the caller function, and address of the indirect argument may be
taken, in either case the indirect function argument is expected to be in default
addr space, not the alloca address space.
Therefore, the indirect function argument should be mapped to the temp var
casted to default address space. The caller will cast it to alloca addr space
when passing it to the callee. In the callee, the argument is also casted to the
default address space and used.
CallArg is refactored to facilitate this fix.
Differential Revision: https://reviews.llvm.org/D34367
llvm-svn: 326946
OpenCL runtime tracks the invoke function emitted for
any block expression. Due to restrictions on blocks in
OpenCL (v2.0 s6.12.5), it is always possible to know the
block invoke function when emitting call of block expression
or __enqueue_kernel builtin functions. Since __enqueu_kernel
already has an argument for the invoke function, it is redundant
to have invoke function member in the llvm block literal structure.
This patch removes invoke function from the llvm block literal
structure. It also removes the bitcast of block invoke function
to the generic block literal type which is useless for OpenCL.
This will save some space for the kernel argument, and also
eliminate some store instructions.
Differential Revision: https://reviews.llvm.org/D43783
llvm-svn: 326937
The tests that failed on a windows host have been fixed.
Original message:
Start setting dso_local for COFF.
With this there are still some GVs where we don't set dso_local
because setGVProperties is never called. I intend to fix that in
followup commits. This is just the bare minimum to teach
shouldAssumeDSOLocal what it should do for COFF.
llvm-svn: 325940
Summary:
OpenCL 2.0 specification defines '-cl-uniform-work-group-size' option,
which requires that the global work-size be a multiple of the work-group
size specified to clEnqueueNDRangeKernel and allows optimizations that
are made possible by this restriction.
The patch introduces the support of this option.
To keep information about whether an OpenCL kernel has uniform work
group size or not, clang generates 'uniform-work-group-size' function
attribute for every kernel:
- "uniform-work-group-size"="true" for OpenCL 1.2 and lower,
- "uniform-work-group-size"="true" for OpenCL 2.0 and higher if
'-cl-uniform-work-group-size' option was specified,
- "uniform-work-group-size"="false" for OpenCL 2.0 and higher if no
'-cl-uniform-work-group-size' options was specified.
If the function is not an OpenCL kernel, 'uniform-work-group-size'
attribute isn't generated.
Patch by: krisb
Reviewers: yaxunl, Anastasia, b-sumner
Reviewed By: yaxunl, Anastasia
Subscribers: nhaehnle, yaxunl, Anastasia, cfe-commits
Differential Revision: https://reviews.llvm.org/D43570
llvm-svn: 325771
The following test case causes issue with codegen of __enqueue_block
void (^block)(void) = ^{ callee(id, out); };
enqueue_kernel(queue, 0, ndrange, block);
Clang first does codegen for block expression in the first line and deletes its block info.
Clang then tries to do codegen for the same block expression again for the second line,
and fails because the block info is gone.
The fix is to do normal codegen for both lines. Introduce an API to OpenCL runtime to
record llvm block invoke function and llvm block literal emitted for each AST block
expression, and use the recorded information for generating the wrapper kernel.
The EmitBlockLiteral APIs are cleaned up to minimize changes to the normal codegen
of blocks.
Another minor issue is that some clean up AST expression is generated for block
with captures, which can be stripped by IgnoreImplicit.
Differential Revision: https://reviews.llvm.org/D43240
llvm-svn: 325264
Summary:
Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset
intrinsics. This change updates the Clang tests for this change.
The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument
which is required to be a constant integer. It represents the alignment of the
dest (and source), and so must be the minimum of the actual alignment of the
two.
This change removes the alignment argument in favour of placing the alignment
attribute on the source and destination pointers of the memory intrinsic call.
For example, code which used to read:
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false)
will now read
call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false)
At this time the source and destination alignments must be the same (Step 1).
Step 2 of the change, to be landed shortly, will relax that contraint and allow
the source and destination to have different alignments.
llvm-svn: 322964
CreateCoercedLoad/CreateCoercedStore assumes pointer argument of
memcpy is in addr space 0, which is not correct and causes invalid
bitcasts for triple amdgcn---amdgiz.
It is fixed by using alloca addr space instead.
Differential Revision: https://reviews.llvm.org/D40806
llvm-svn: 320000
Summary:
Constant samplers are handled as static variables and clang's code generation
library, which leads to llvm::unreachable. We bypass emitting sampler variable
as static since it's translated to a function call later.
Reviewers: yaxunl, Anastasia
Reviewed By: yaxunl, Anastasia
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D34342
llvm-svn: 318290
Builder save/restores insertion pointer when emitting addr space cast
for alloca, but does not save/restore debug loc, which causes verifier
failure for certain call instructions.
This patch fixes that.
Differential Revision: https://reviews.llvm.org/D39069
llvm-svn: 316484
Matt Arsenault