The problem with the old scheme is that we would need to keep track of
the "next region" and reset the num_threads value after it. The new RT
doesn't do it and an assertion is triggered. The old RT doesn't do it
either, I haven't tested it but I assume a num_threads clause might
impact multiple parallel regions "accidentally". Further, in SPMD mode
num_threads was simply ignored, for some reason beyond me.
In any case, parallel_51 is designed to take the clause value directly,
so let's do that instead.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D113623
The existing CGOpenMPRuntimeAMDGCN and CGOpenMPRuntimeNVPTX classes are
just code bloat. By removing them, the codebase gets a bit cleaner.
Reviewed By: jdoerfert, JonChesterfield, tianshilei1992
Differential Revision: https://reviews.llvm.org/D113421
The existing CGOpenMPRuntimeAMDGCN and CGOpenMPRuntimeNVPTX classes are
just code bloat. By removing them, the codebase gets a bit cleaner.
Reviewed By: jdoerfert, JonChesterfield, tianshilei1992
Differential Revision: https://reviews.llvm.org/D113421
Sequel patch to https://reviews.llvm.org/D111293.
Remove call to CodeGenFunction::InitTempAlloca() from OpenMP related
codegen part.
Also remove the metadata `!llvm.access.group` from the updated lit
tests.
Reviewed By: rjmccall
Differential Revision: https://reviews.llvm.org/D111316
This patch adds support for the
`__kmpc_get_hardware_num_threads_in_block` function that returns the
number of threads. This was missing in the new runtime and was used by
the AMDGPU plugin which prevented it from using the new runtime. This
patchs also unified the interface for getting the thread numbers in the
frontend.
Originally authored by jdoerfert.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D111475
This patch adds two flags to be supported for the new runtime. The flags
are `-fopenmp-assume-threads-oversubscription` and
-fopenmp-assume-teams-oversubscription`. These add global values that
can be checked by the work sharing runtime functions to make better
judgements about how to distribute work between the threads.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D111348
The execution mode of a kernel is stored in a global variable, whose value means:
- 0 - SPMD mode
- 1 - indicates generic mode
- 2 - SPMD mode execution with generic mode semantics
We are going to add support for SIMD execution mode. It will be come with another
execution mode, such as SIMD-generic mode. As a result, this value-based indicator
is not flexible.
This patch changes to bitset based solution to encode execution mode. Each
position is:
[0] - generic mode
[1] - SPMD mode
[2] - SIMD mode (will be added later)
In this way, `0x1` is generic mode, `0x2` is SPMD mode, and `0x3` is SPMD mode
execution with generic mode semantics. In the future after we add the support for
SIMD mode, `0b1xx` will be in SIMD mode.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D110029
Parallel regions are outlined as functions with capture variables explicitly generated as distinct parameters in the function's argument list. That complicates the fork_call interface in the OpenMP runtime: (1) the fork_call is variadic since there is a variable number of arguments to forward to the outlined function, (2) wrapping/unwrapping arguments happens in the OpenMP runtime, which is sub-optimal, has been a source of ABI bugs, and has a hardcoded limit (16) in the number of arguments, (3) forwarded arguments must cast to pointer types, which complicates debugging. This patch avoids those issues by aggregating captured arguments in a struct to pass to the fork_call.
Reviewed By: jdoerfert, jhuber6
Differential Revision: https://reviews.llvm.org/D102107
This patch introduces the flags `-fopenmp-target-debug` and
`-fopenmp-target-debug=` to set the value of a global in the device.
This will be used to enable or disable debugging features statically in
the device runtime library.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D109544
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
[nfc] Replaces enum indices into an array with a struct. Named the
fields to match the enum, leaves memory layout and initialization unchanged.
Motivation is to later safely remove dead fields and replace redundant ones
with (compile time) computation. It should also be possible to factor some
common fields into a base and introduce a gfx10 amdgpu instance with less
duplication than the arrays of integers require.
Reviewed By: ronlieb
Differential Revision: https://reviews.llvm.org/D108339
This patch changes `__kmpc_free_shared` to take an additional argument
corresponding to the associated allocation's size. This makes it easier to
implement the allocator in the runtime.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106496
Remove uses of to-be-deprecated API. In cases where the correct
element type was not immediately obvious to me, fall back to
explicit getPointerElementType().
Parallel regions are outlined as functions with capture variables explicitly generated as distinct parameters in the function's argument list. That complicates the fork_call interface in the OpenMP runtime: (1) the fork_call is variadic since there is a variable number of arguments to forward to the outlined function, (2) wrapping/unwrapping arguments happens in the OpenMP runtime, which is sub-optimal, has been a source of ABI bugs, and has a hardcoded limit (16) in the number of arguments, (3) forwarded arguments must cast to pointer types, which complicates debugging. This patch avoids those issues by aggregating captured arguments in a struct to pass to the fork_call.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D102107
In the spirit of TRegions [0], this patch provides a simpler and uniform
interface for a kernel to set up the device runtime. The OMPIRBuilder is
used for reuse in Flang. A custom state machine will be generated in the
follow up patch.
The "surplus" threads of the "master warp" will not exit early anymore
so we need to use non-aligned barriers. The new runtime will not have an
extra warp but also require these non-aligned barriers.
[0] https://link.springer.com/chapter/10.1007/978-3-030-28596-8_11
This was in parts extracted from D59319.
Reviewed By: ABataev, JonChesterfield
Differential Revision: https://reviews.llvm.org/D101976
Broke check-clang, see https://reviews.llvm.org/D102307#2869065
Ran `git revert -n ebbe149a6f08535ede848a531a601ae6591cfbc5..269416d41908bb670f67af689155d5ab8eea689a`
In the spirit of TRegions [0], this patch provides a simpler and uniform
interface for a kernel to set up the device runtime. The OMPIRBuilder is
used for reuse in Flang. A custom state machine will be generated in the
follow up patch.
The "surplus" threads of the "master warp" will not exit early anymore
so we need to use non-aligned barriers. The new runtime will not have an
extra warp but also require these non-aligned barriers.
[0] https://link.springer.com/chapter/10.1007/978-3-030-28596-8_11
This was in parts extracted from D59319.
Reviewed By: ABataev, JonChesterfield
Differential Revision: https://reviews.llvm.org/D101976
Summary:
Memory globalization is required to maintain OpenMP standard semantics for data sharing between
worker and master threads. The GPU cannot share data between its threads so must allocate global or
shared memory to store the data in. Currently this is implemented fully in the frontend using the
`__kmpc_data_sharing_push_stack` and __kmpc_data_sharing_pop_stack` functions to emulate standard
CPU stack sharing. The front-end scans the target region for variables that escape the region and
must be shared between the threads. Each variable then has a field created for it in a global record
type.
This patch replaces this functinality with a single allocation command, effectively mimicing an
alloca instruction for the variables that must be shared between the threads. This will be much
slower than the current solution, but makes it much easier to optimize as we can analyze each
variable independently and determine if it is not captured. In the future, we can replace these
calls with an `alloca` and small allocations can be pushed to shared memory.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D97680
This revision simplifies Clang codegen for parallel regions in OpenMP GPU target offloading and corresponding changes in libomptarget: SPMD/non-SPMD parallel calls are unified under a single `kmpc_parallel_51` runtime entry point for parallel regions (which will be commonized between target, host-side parallel regions), data sharing is internalized to the runtime. Tests have been auto-generated using `update_cc_test_checks.py`. Also, the revision contains changes to OpenMPOpt for remark creation on target offloading regions.
Reviewed By: jdoerfert, Meinersbur
Differential Revision: https://reviews.llvm.org/D95976
The `noinline` for non-SPMD parallel functions is probably not necessary
but as long as we use it we should put it on the outermost parallel
function, which is the wrapper, not the actual outlined function.
Resolves PR49752
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D99506
This removes some (but not all) uses of type-less CreateGEP()
and CreateInBoundsGEP() APIs, which are incompatible with opaque
pointers.
There are a still a number of tricky uses left, as well as many
more variation APIs for CreateGEP.
The patch only plumbs through the option necessary for targeting sm_86 GPUs w/o
adding any new functionality.
Differential Revision: https://reviews.llvm.org/D95974
Whenever we enter a new OpenMP data environment we want to enter a
function to simplify reasoning. Later we probably want to remove the
entire specialization wrt. the if clause and pass the result to the
runtime, for now this should fix PR48686.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D94315
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
[libomptarget][nvptx] Undef, weak shared variables
Shared variables on nvptx, and LDS on amdgcn, are uninitialized at
the start of kernel execution. Therefore create the variables with
undef instead of zeros, motivated in part by the amdgcn back end
rejecting LDS+initializer.
Common is zero initialized, which seems incompatible with shared. Thus
change them to weak, following the direction of
https://reviews.llvm.org/rG7b3eabdcd215
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D90248
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
Summary:
Replace the OpenMP Runtime Library functions used in CGOpenMPRuntimeGPU
for OpenMP device code generation with ones in OMPKinds.def and use
OMPIRBuilder for generating runtime calls. This allows us to
consolidate more OpenMP code generation into the OMPIRBuilder. Future
additions to the GPU runtime functions should now go in OMPKinds.def
Reviewers: jdoerfert
Subscribers: aaron.ballman cfe-commits guansong llvm-commits sstefan1 yaxunl
Tags: #OpenMP #LLVM #clang
Differential Revision: https://reviews.llvm.org/D88430
Currently CUDA/HIP toolchain uses "unknown" as bound arch
for offload action for fat binary. This causes -mcpu or -march
with "unknown" added in HIPToolChain::TranslateArgs or
CUDAToolChain::TranslateArgs.
This causes issue for https://reviews.llvm.org/D88377 since
HIP toolchain needs to check -mcpu in HIPToolChain::TranslateArgs.
The bound arch of offload action for fat binary is not really
used, therefore set it to CudaArch::UNUSED.
Differential Revision: https://reviews.llvm.org/D88524
Failing tests on Arm due to the tests automatically populating
incomatible pointer width architectures. Reverting until the tests are
updated. Failing tests:
OpenMP/distribute_parallel_for_num_threads_codegen.cpp
OpenMP/distribute_parallel_for_if_codegen.cpp
OpenMP/distribute_parallel_for_simd_if_codegen.cpp
OpenMP/distribute_parallel_for_simd_num_threads_codegen.cpp
OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp
OpenMP/target_teams_distribute_parallel_for_simd_if_codegen.cpp
OpenMP/teams_distribute_parallel_for_if_codegen.cpp
OpenMP/teams_distribute_parallel_for_simd_if_codegen.cpp
This reverts commit 90eaedda9b8ef46e2c0c1b8bce33e98a3adbb68c.
Summary:
Replace the OpenMP Runtime Library functions used in CGOpenMPRuntimeGPU
for OpenMP device code generation with ones in OMPKinds.def and use
OMPIRBuilder for generating runtime calls. This allows us to consolidate
more OpenMP code generation into the OMPIRBuilder. This patch also
invalidates specifying target architectures with conflicting pointer
sizes.
Reviewers: jdoerfert
Subscribers: aaron.ballman cfe-commits guansong llvm-commits sstefan1 yaxunl
Tags: #OpenMP #Clang #LLVM
Differential Revision: https://reviews.llvm.org/D88430
When we implement OpenMP GPU reductions we use type punning a lot during
the shuffle and reduce operations. This is not always compatible with
language rules on aliasing. So far we generated TBAA which later allowed
to remove some of the reduce code as accesses and initialization were
"known to not alias". With this patch we avoid TBAA in this step,
hopefully for all accesses that we need to.
Verified on the reproducer of PR46156 and QMCPack.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D86037
We can use this to remove some calls to initFeatureMap from Sema
and CodeGen when a function doesn't have a target attribute.
This reduces compile time of the linux kernel where this map
is needed to diagnose some inline assembly constraints based
on whether sse, avx, or avx512 is enabled.
Differential Revision: https://reviews.llvm.org/D85807
