Use the in-project clang, llvm-link and opt if available and unless
CMake cache variables specify to use a different compiler. This applies
D101265 to the new DeviceRTL's CMakeLists.txt which was copied before
D101265 was applied.
Fixes the openmp-offloading-cuda-runtime builder which was failing
since D110006.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D110251
Store queues in unique_ptr so they are destroyed when the global DeviceInfo is. Currently they leak which raises an assert in debug builds of hsa.
Reviewed By: pdhaliwal
Differential Revision: https://reviews.llvm.org/D109511
This patch fixes a data-race observed when using the new device runtime
library. The Internal control variable for the parallel level is read in
the `__kmpc_parallel_51` function while it could potentially be written
by other threads. This causes data corruption and will cause
nondetermistic behaviour in the runtime. This patch fixes this by adding
an explicit synchronization before the region starts.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D110366
This is a follow-up of D110029, which uses bitset to indicate execution mode. This patches makes the changes in the function call.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D110279
This patch adds support for an RAII struct that will print function
traces when placed inside of a function declaration. Each successive
call will increase the indentation to make it easier to visually
inspect.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D110202
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
Summary:
The thread ID function was reintroduced in D110195, but could
potentially be removed by the optimizer. Make the function noinline to
preserve the call sites and add it to the externalization RAII so its
definition is not removed by the attributor.
The new device runtime library currently lacks the
`kmpc_get_hardware_thread_id_in_block` function which is currently used
when doing the SPMDzation optimization. This call would be introduced
through the optimization and then cause a linking error because it was
not present. This patch adds support for this runtime call.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D110195
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 adds support for using dynamic shared memory in the new
device runtime. The new function `__kmpc_get_dynamic_shared` will return a
pointer to the buffer of dynamic shared memory. Currently the amount of memory
allocated is set by an environment variable.
In the future this amount will be added to the amount used for the smart stack
which will be configured in a similar way.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D110006
This patch adds fields for the device number and number of devices into
the device environment struct and debugging values.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D110004
This patch implements the `__assert_fail` function in the new device
runtime. This allows users and developers to use the standars assert
function inside of the device.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D109886
The defintion of OFFLOAD_SUCCESS and OFFLOAD_FAIL used in plugin APIs and libomptarget public APIs are not consistent.
Create __tgt_target_return_t for libomptarget public APIs.
Differential Revision: https://reviews.llvm.org/D109304
The hsa library must be initialized before any calls into it and
destructed after the last call into it. There have been a number of bugs in
this area related to member variables which would like to use raii to manage
resources acquired from hsa.
This patch moves the init/shutdown of hsa into a class, such that when used as
the first member variable (could be a base), the lifetime of other member
variables are reliably scoped within it. This will allow other classes to use
raii reliably when used as member variables within the global.
Reviewed By: pdhaliwal
Differential Revision: https://reviews.llvm.org/D109512
Given D109057, change test runner to use the libomptarget-x-bc-path
argument instead of the LIBRARY_PATH environment variable to find the device
library.
Also drop the use of LIBRARY_PATH environment variable as it is far
too easy to pull in the device library from an unrelated toolchain by accident
with the current setup. No loss in flexibility to developers as the clang
commandline used here is still available.
Reviewed By: jdoerfert, tianshilei1992
Differential Revision: https://reviews.llvm.org/D109061
Using std::vector<DeviceTy> requires implementing copy constructor and copied assign operator for DeviceTy.
Indeed DeviceTy should never be copied. After changing to std::vector<std::unique_ptr<DeviceTy>>,
All the unsafe copy constructor and copy assign operator implementations can be removed.
Compilers mark them deleted due to mutex or underlying objects and this is the desired behavior.
Differential Revision: https://reviews.llvm.org/D109276
Use the same debug print as the rest of libomptarget plugins with
the same environment control. Also drop the max queue size debugging hook as
I don't believe it is still in use, can bring it back near the rest of the env
handling in rtl.cpp if someone objects.
That makes most of rt.h and all of utils.cpp unused. Clean that up and simplify
control flow in a couple of places.
Behaviour change is that debug prints that used to use the old environment
variable now use the new one and print in slightly different format, and the
removal of the max queue size variable.
Reviewed By: pdhaliwal
Differential Revision: https://reviews.llvm.org/D108784
Use unique_ptr to achieve the effect of mutable.
Remove mutable keyword of DynRefCount and HoldRefCount
Remove std::shared_ptr from UpdateMtx
Reviewed By: tianshilei1992, grokos
Differential Revision: https://reviews.llvm.org/D109007
As started in D107925, this patch replaces the remaining occurrences
of `UNIFIED_SHARED_MEMORY && TgtPtrBegin == HstPtrBegin` in
`omptarget.cpp` with `IsHostPtr`. The former condition is broken in
the rare case that the device and host happen to use the same address
for their mapped allocations. I don't know how to write a test that's
likely to reveal this case.
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D107928
As discussed in D105990, without this patch, `targetDataBegin`
determines whether to transfer data (as opposed to assuming it's in
shared memory) using the condition `!UseUSM || HasCloseModifier`.
However, this condition is broken if use of discrete memory was forced
by `omp_target_associate_ptr`. This patch extends
`unified_shared_memory/associate_ptr.c` to reveal this case, and it
fixes it using `!IsHostPtr` in `DeviceTy::getTargetPointer` to replace
this condition.
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D107927
This patch is based on comments in D105990. It is NFC according to
the following observations:
1. `CopyMember` is computed as `!IsHostPtr && IsLast`.
2. `DelEntry` is true only if `IsLast` is true.
We apply those observations in order:
```
if ((DelEntry || Always || CopyMember) && !IsHostPtr)
if ((DelEntry || Always || IsLast) && !IsHostPtr)
if ((Always || IsLast) && !IsHostPtr)
```
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D107926
As discussed in D105990, without this patch, `targetDataEnd`
determines whether to transfer data or delete a device mapping (as
opposed to assuming it's in shared memory) using two different
conditions, each of which is broken for some cases:
1. `!(UNIFIED_SHARED_MEMORY && TgtPtrBegin == HstPtrBegin)`: The
broken case is rare: the device and host might happen to use the
same address for their mapped allocations. I don't know how to
write a test that's likely to reveal this case, but this patch does
fix it, as discussed below.
2. `!UNIFIED_SHARED_MEMORY || HasCloseModifier`: There are at least
two broken cases:
1. The `close` modifier might have been specified on an `omp
target enter data` but not the corresponding `omp target exit
data`, which thus might falsely assume a mapping is in shared
memory. The test `unified_shared_memory/close_enter_exit.c`
already has a missing deletion as a result, and this patch adds
a check for that. This patch also adds the new test
`close_member.c` to reveal a missing transfer and deletion.
2. Use of discrete memory might have been forced by
`omp_target_associate_ptr`, as in the test
`unified_shared_memory/api.c`. In the current `targetDataEnd`
implementation, this condition turns out not be used for this
case: because the reference count is infinite, a transfer is
possible only with an `always` modifier, and this condition is
never used in that case. To ensure it's never used for that
case in the future, this patch adds the test
`unified_shared_memory/associate_ptr.c`.
Fortunately, `DeviceTy::getTgtPtrBegin` already has a solution: it
reports whether the allocation was found in shared memory via the
variable `IsHostPtr`.
After this patch, `HasCloseModifier` is no longer used in
`targetDataEnd`, and I wonder if the `close` modifier is ever useful
on an `omp target data end`.
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D107925
Given D109057, change test runner to use the libomptarget-x-bc-path
argument instead of the LIBRARY_PATH environment variable to find the device
library.
Also drop the use of LIBRARY_PATH environment variable as it is far
too easy to pull in the device library from an unrelated toolchain by accident
with the current setup. No loss in flexibility to developers as the clang
commandline used here is still available.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D109061
Using rpath instead of LD_LIBRARY_PATH to find libomp.so and
libomptarget.so lets one rerun the already built test executables without
setting environment variables and removes the risk of the test runner picking
up different libraries to the developer debugging the failure.
rpath usually means runpath, which is not transitive, so set runpath on
libomptarget itself so that it can find the plugins located next to it,
spelled $ORIGIN. This provides sufficient functionality to drop D102043
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D109071
This patch implements OpenMP runtime support for an original OpenMP
extension we have developed to support OpenACC: the `ompx_hold` map
type modifier. The previous patch in this series, D106509, implements
Clang support and documents the new functionality in detail.
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D106510
In some build configurations, the target we depend on is not available for declaring the build dependency.
We only need to declare the build dependency, if the build target is available in the same build.
Fixes the issue raised in https://reviews.llvm.org/D107156#2969862
This patch should go into release/13 together with D108404
Differential Revision: https://reviews.llvm.org/D108868
`CU_EVENT_DEFAULT` is defined in CUDA header. It should be added to
`openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.h` for CUDA free build.
Reviewed By: ronlieb
Differential Revision: https://reviews.llvm.org/D108878
This patch adds the support form event related interfaces, which will be used
later to fix data race. See D104418 for more details.
Reviewed By: jdoerfert, ye-luo
Differential Revision: https://reviews.llvm.org/D108528
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
Lets the amdgpu plugin write to omptarget_device_environment
to enable debugging. Intend to use in the near future to record the
wavesize that a given deviceRTL was compiled with for running on hardware
that supports 32 or 64.
Patch sets all the attributes that are useful. Notably .data means the variable
is set by writing to host memory before copying to the GPU instead of launching
a kernel to update the image. Can simplify the plugin slightly to drop the
code for patching after load if this is used consistently.
NFC on nvptx, cuda plugin seems to work fine without any annotations.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D108698
Move most debug printing in rtl.cpp behind DP() macro
Adjust the print output for gpu arch mismatch when the architectures match
Convert an assert into graceful failure
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D108562
The use of `$<TARGET_FILE:clang>` was adapted too broadly from D101265.
Fixes llvm.org/PR51579
Also see discussion in D108534.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D108640
With uses of g_atl_machine gone, a significant portion of dead
code has been removed.
This patch depends on D104691 and D104695.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D104696
Compiled nvptx devicertl as freestanding, breaking the
dependency on host glibc and gcc-multilibs. Thus build it by default.
Comes at the cost of #defining out printf. Tried mapping it onto
__builtin_printf but that gets transformed back to printf instead
of hitting the cuda/openmp lowering transform.
Printf could be preserved by one of:
- dropping all the standard headers and ffreestanding
- providing a header only printf implementation
- changing the compiler handling of printf
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D108349
Add include path to the cmakefiles and set the target_impl enums
from the llvm constants instead of copying the values.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D108391
D107156 and D107320 are not sufficient when OpenMP is built as llvm runtime
(LLVM_ENABLE_RUNTIMES=openmp) because dependencies only work within the same
cmake instance.
We could limit the dependency to cases where libomptarget/plugins are really
built. But compared to the whole llvm project, building openmp runtime is
negligible and postponing the build of OpenMP runtime after the dependencies
are ready seems reasonable.
The direct dependency introduced in D107156 and D107320 is necessary for the
case where OpenMP is built as llvm project (LLVM_ENABLE_PROJECTS=openmp).
Differential Revision: https://reviews.llvm.org/D108404
A new rule is added in 5.0:
If a list item appears in a reduction, lastprivate or linear clause
on a combined target construct then it is treated as if it also appears
in a map clause with a map-type of tofrom.
Currently map clauses for all capture variables are added implicitly.
But missing for list item of expression for array elements or array
sections.
The change is to add implicit map clause for array of elements used in
reduction clause. Skip adding map clause if the expression is not
mappable.
Noted: For linear and lastprivate, since only variable name is
accepted, the map has been added though capture variables.
To do so:
During the mappable checking, if error, ignore diagnose and skip
adding implicit map clause.
The changes:
1> Add code to generate implicit map in ActOnOpenMPExecutableDirective,
for omp 5.0 and up.
2> Add extra default parameter NoDiagnose in ActOnOpenMPMapClause:
Use that to skip error as well as skip adding implicit map during the
mappable checking.
Note: there are only tow places need to be check for NoDiagnose. Rest
of them either the check is for < omp 5.0 or the error already generated for
reduction clause.
Differential Revision: https://reviews.llvm.org/D108132
[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
Use uint64_t for lanemask on all GPU architectures at the interface
with clang. Updates tests. The deviceRTL is always linked as IR so the zext
and trunc introduced for wave32 architectures will fold after inlining.
Simplification partly motivated by amdgpu gfx10 which will be wave32 and
is awkward to express in the current arch-dependant typedef interface.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D108317
Currently, the runtime returns an error when the `exec_mode` global is
not present. The expected behvaiour is that the region will default to
Generic. This prevents global constructors from being called because
they do not contain execution mode globals.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D108255
Removed redundant assignment from condition which causes gcc to emit the following error:
error: operation on ‘MoveData’ may be undefined [-Werror=sequence-point]
targetDataEnd and targetDataBegin compute CopyMember/copy differently,
and I don't see why they should. This patch eliminates one of those
differences by making a simplifying NFC change to targetDataEnd.
The change is NFC as follows. The change only affects the case when
`!UNIFIED_SHARED_MEMORY || HasCloseModifier`. In that case, the
following points are always true:
* The value of CopyMember is relevant later only if DelEntry = false.
* DelEntry = false only if one of the following is true:
* IsLast = false. In this case, it's always true that CopyMember
= false = IsLast.
* `MEMBER_OF && !PTR_AND_OBJ` is true. In this case, CopyMember =
IsLast.
* Thus, if CopyMember is relevant, CopyMember = IsLast.
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D105990
On FreeBSD, the `environ` symbol is undefined at link time for shared
libraries, but resolved by the dynamic linker at runtime. Therefore,
allow the symbol to be undefined when creating a shared library, by
using the `--allow-shlib-undefined` linker flag, instead of `-z defs`
(a.k.a `--no-undefined`).
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D107698
On FreeBSD, the system `<libelf.h>` already declares `struct Elf_Note`
indirectly (via `<sys/elf_common.h>`). This results in compile errors
when building the libomptarget amdgpu plugin. Avoid redeclaring `struct
Elf_Note` on FreeBSD to fix the errors.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D107661
When loading libomptarget, the init function in libomptarget/src/rtl.cpp
will search for the libomptarget_start_tool function using libdl.
libomptarget_start_tool will pass those OMPT callbacks related to target
constructs to libomptarget
Differential Revision: https://reviews.llvm.org/D99803
Remove --cuda-path=CUDA_TOOLKIT_ROOT_DIR-NOTFOUND
from the invocation of non-nvptx test cases. Better signal
to noise ratio on other architectures.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D107074
When using `-DLLVM_ENABLED_RUNTIMES` instead of `-DLLVM_ENABLED_PROJECTS`
the `llvm-omp-device-info` tool is not compiled or installed.
In general, no llvm tool would be build on runtimes, because the
-DLLVM_BUILD_TOOLS flag is removed by the way runtimes compilation calls
cmake again.
This patch is simple. Just forward the value of this flag to the
runtime cmake command.
I'm also removing an unnecessary comment in the compilation of the tool
Differential Revision: https://reviews.llvm.org/D107177
All `nowait` series of interfaces in `libomptarget` accept four more arguments (`int32_t depNum, void *depList, int32_t noAliasDepNum, void *noAliasDepList`) compared with their counterparts w/o `nowait`. These extra arguments were expected for dependence resolution, potentially lowered to device side. Current implementation calls `libomp` function `__kmpc_omp_taskwait`. However, the front end simply ignores them, that these four arguments are not emitted at all. As a consequence, the `depNum` and `noAliasDepNum` are garbage, which could lead to unnecessary task wait.
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D107164
This patch fixes the error reported in D106751. When there is no CUDA SDK
installed in the system, the build fails due to missing `CU_DEVICE_ATTRIBUTE`
variables.
Using @zsrkmyn sugested fix
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106933
This patch introduces the `llvm-omp-device-info` tool, which uses the
omptarget library and interface to query the device info from all the
available devices as seen by OpenMP. This is inspired by PGI's `pgaccelinfo`
Since omptarget usually requires a description structure with executable
kernels, I split the initialization of the RTLs and Devices to be able to
initialize all possible devices and query each of them.
This revision relies on the patch that introduces the print device info.
A limitation is that the order in which the devices are initialized, and the
corresponding device ID is not necesarily the one seen by OpenMP.
The changes are as follows:
1. Separate the RTL initialization that was performed in `RegisterLib` to its own `initRTLonce` function
2. Create an `initAllRTLs` method that initializes all available RTLs at runtime
3. Created the `llvm-deviceinfo.cpp` tool that uses `omptarget` to query each device and prints its information.
Example Output:
```
Device (0):
print_device_info not implemented
Device (1):
print_device_info not implemented
Device (2):
print_device_info not implemented
Device (3):
print_device_info not implemented
Device (4):
CUDA Driver Version: 11000
CUDA Device Number: 0
Device Name: Quadro P1000
Global Memory Size: 4236312576 bytes
Number of Multiprocessors: 5
Concurrent Copy and Execution: Yes
Total Constant Memory: 65536 bytes
Max Shared Memory per Block: 49152 bytes
Registers per Block: 65536
Warp Size: 32 Threads
Maximum Threads per Block: 1024
Maximum Block Dimensions: 1024, 1024, 64
Maximum Grid Dimensions: 2147483647 x 65535 x 65535
Maximum Memory Pitch: 2147483647 bytes
Texture Alignment: 512 bytes
Clock Rate: 1480500 kHz
Execution Timeout: Yes
Integrated Device: No
Can Map Host Memory: Yes
Compute Mode: DEFAULT
Concurrent Kernels: Yes
ECC Enabled: No
Memory Clock Rate: 2505000 kHz
Memory Bus Width: 128 bits
L2 Cache Size: 1048576 bytes
Max Threads Per SMP: 2048
Async Engines: Yes (2)
Unified Addressing: Yes
Managed Memory: Yes
Concurrent Managed Memory: Yes
Preemption Supported: Yes
Cooperative Launch: Yes
Multi-Device Boars: No
Compute Capabilities: 61
```
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D106752
This patch introduces a function in the device's plugin to print the
device information. This patch relates to another patch that introduces
a CLI tool to obtain the device information from the omplibrary directly.
It is inspired by PGI's pgaccelinfo.
The modifications are as follows:
1. Introduce the optional `void __tgt_rtl_print_device_info(RTLdevID)` function into the RTL.
2. Introduce the `bool __tgt_print_device_info(devID)` function into `omptarget` interface. Returns false if the RTL is not implemented
3. Added `bool printDeviceInfo(RTLDevID)` to the `DeviceTy`
4. Implement the `__tgt_rtl_print_device_info` for CUDA. Added additional CUDA Runtime calls.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106751
The device runtime contains several calls to `__kmpc_get_hardware_num_threads_in_block`
and `__kmpc_get_hardware_num_blocks`. If the thread_limit and the num_teams are constant,
these calls can be folded to the constant value.
In this patch we use the already introduced `AAFoldRuntimeCall` and the `NumTeams` and
`NumThreads` kernel attributes (to be introduced in a different patch) to fold these functions.
The code checks all the kernels, and if their attributes match, the functions are folded.
In the future we will explore specializing for multiple values of NumThreads and NumTeams.
Depends on D106390
Reviewed By: jdoerfert, JonChesterfield
Differential Revision: https://reviews.llvm.org/D106033
The new method of sharing variables introduces a `__kmpc_alloc_shared` call
that cannot be removed in the middle end because of its non-constant argument
and unconnected free. This patch reverts this to the old method that used a
static amount of shared memory for sharing variables.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106905
The "old" OpenMP GPU device runtime (D14254) has served us well for many
years but modernizing it has caused some pain recently. This patch
introduces an alternative which is mostly written from scratch embracing
OpenMP 5.X, C++, LLVM coding style (where applicable), and conceptual
interfaces. This new runtime is opt-in through a clang flag (D106793).
The new runtime is currently only build for nvptx and has "-new" in its
name.
The design is tailored towards middle-end optimizations rather than
front-end code generation choices, a trend we already started in the old
runtime a while back. In contrast to the old one, state is organized in
a simple manner rather than a "smart" one. While this can induce costs
it helps optimizations. Our expectation is that the majority of codes
can be optimized and a "simple" design is therefore preferable. The new
runtime does also avoid users to pay for things they do not use,
especially wrt. memory. The unlikely case of nested parallelism is
supported but costly to make the more likely case use less resources.
The worksharing and reduction implementation have been taken from the
old runtime and will be rewritten in the future if necessary.
Documentation and debug features are still mostly missing and will be
added over time.
All external symbols start with `__kmpc` for legacy reasons but should
be renamed once we switch over to a single runtime. All internal symbols
are placed in appropriate namespaces (anonymous or `_OMP`) to avoid name
clashes with user symbols.
Differential Revision: https://reviews.llvm.org/D106803
Similar to D105787, this patch tries to fold `__kmpc_parallel_level` if possible.
Note that `__kmpc_parallel_level` doesn't take activeness into consideration,
based on current `deviceRTLs`, its return value can be such as 0, 1, 2, instead
of 0, 129, 130, etc. that also indicate activeness.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106154
Default to building the amdgpu plugin to use dlopen when hsa is
not found instead of disabling it.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106600
If hsa_init fails, subsequent calls into hsa are not safe. Except for
hsa_init, but we don't retry on failure.
This patch:
- deletes a print that called into hsa to ask why it can't call into hsa
- drops a merge conflict block next to that print
- reliably initializes number of devices to zero
- skips the plugin destructor contents if the constructor failed to init hsa
Tested by making hsa_init return error, and by forcing the dynamic library
use which was then deleted from disk. Before this patch, both segv. After it,
friendly message about offloading being unavailable.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106774
Default to building the amdgpu plugin to use dlopen when hsa is
not found instead of disabling it.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106600
We build `deviceRTLs` with `-O1` by default, which also triggers OpenMPOpt. When
the info cache is created, some attributes are removed. As a result, although we
mark a few functions `noinline`, they are still inlined when the bitcode library
is generated. This can cause an issue in middle end optimization.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106710
Bug 50022 [0] reports target nowait fails in certain case, which is added in this
patch. The root cause of the failure is, when the second task is created, its
parent's `td_incomplete_child_tasks` will not be incremented because there is no
parallel region here thus its team is serialized. Therefore, when the initial
thread is waiting for its unfinished children tasks, it thought there is only
one, the first task, because it is hidden helper task, so it is tracked. The
second task will only be pushed to the queue when the first task is finished.
However, when the first task finishes, it first decrements the counter of its
parent, and then release dependences. Once the counter is decremented, the thread
will move on because its counter is reset, but actually, the second task has not
been executed at all. As a result, since in this case, the main function finishes,
then `libomp` starts to destroy. When the second task is pushed somewhere, all
some of the structures might already have already been destroyed, then anything
could happen.
This patch simply moves `__kmp_release_deps` ahead of decrement of the counter.
In this way, we can make sure that the initial thread is aware of the existence
of another task(s) so it will not move on. In addition, in order to tackle
dependence chain starting with hidden helper thread, when hidden helper task is
encountered, we force the task to release dependences.
Reference:
[0] https://bugs.llvm.org/show_bug.cgi?id=50022
Reviewed By: AndreyChurbanov
Differential Revision: https://reviews.llvm.org/D106519
Unrolling this loop provides better performance in practice because it is
executed on the device and is likely to be very small.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D106692
This patch tries to partially fix one of the two data race issues reported in
[1] by introducing a per-entry mutex. Additional discussion can also be found in
D104418, which will also be refined to fix another data race problem.
Here is how it works. Like before, `DataMapMtx` is still being used for mapping
table lookup and update. In any case, we will get a table entry. If we need to
make a data transfer (update the data on the device), we need to lock the entry
right before releasing `DataMapMtx`, and the issue of data transfer should be
after releasing `DataMapMtx`, and the entry is unlocked afterwards. This can
guarantee that: 1) issue of data movement is not in critical region, which will
not affect performance too much, and also will not affect other threads that don't
touch the same entry; 2) if another thread accesses the same entry, the state of
data movement is consistent (which requires that a thread must first get the
update lock before getting data movement information).
For a target that doesn't support async data transfer, issue of data movement is
data transfer. This two-lock design can potentially improve concurrency compared
with the design that guards data movement with `DataMapMtx` as well. For a target
that supports async data movement, we could simply attach the event between the
issue of data movement and unlock the entry. For a thread that wants to get the
event, it must first get the lock. This can also get rid of the busy wait until
the event pointer is valid.
Reference:
[1] https://bugs.llvm.org/show_bug.cgi?id=49940
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D104555
With D106496 we can make the globalization fallback stack much simpler
and this version doesn't seem to experience the spurious failures and
deadlocks we have seen before.
Differential Revision: https://reviews.llvm.org/D106576
This patch adds support for two environment variables to configure the device.
``LIBOMPTARGET_STACK_SIZE`` sets the amount of memory in bytes that each thread
has for its stack. ``LIBOMPTARGET_HEAP_SIZE`` sets the amount of heap memory
that can be allocated using malloc / free on the device.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106627
These functions should follow the camel case convention. These are really easy to change
and are needed for D106033.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D106390
AMDGPU can assume Elf64 so doesn't need to abstract over Elf32
Drop a few other unused headers at the same time. Now only llvm elf
and libelf are used by the plugin.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106579
AMDGPU plugin equivalent of D95155, build without HSA installed locally
Compiles a new file, plugins/amdgpu/dynamic_hsa/hsa.cpp, to an object file that
exposes the same symbols that the plugin presently uses from hsa. The object
file contains dlopen of hsa and cached dlsym calls. Also provides header files
corresponding to the subset that is used.
This is behind a feature flag, LIBOMPTARGET_FORCE_DLOPEN_LIBHSA, default off.
That allows developers to build against the dlopen/dlsym implementation, e.g.
while testing this mode.
Enabling by default will cause this plugin to build on a wider variety of
machines than it does at present so may break some CI builds. That risk can
be minimised by reviewing the header dependencies of the library and ensuring
it doesn't use any libraries that are not already used by libomptarget.
Separating the implementation from enabling by default in case the latter needs
to be rolled back after wider CI results.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106559
Revision of D102858. Raise dlwrap arity argument to template argument
so the correct value is given in the error message. E.g. '2 == 1' instead of
'2 == trait<>::nargs'.
Arity higher than it should be:
Before diff
```
$/plugins/cuda/dynamic_cuda/cuda.cpp:23:1: error:
static_assert failed due to requirement '2 == trait<cudaError_enum (*)(unsigned int)>::nargs'
"Arity Error"
DLWRAP_INTERNAL(cuInit, 2);
^~~~~~~~~~~~~~~~~~~~~~~~~~
...
$/include/dlwrap.h:166:3: note: expanded from macro
'DLWRAP_COMMON'
static_assert(ARITY == trait<decltype(&SYMBOL)>::nargs, "Arity Error"); \
```
After diff
In file included from $/plugins/cuda/dynamic_cuda/cuda.cpp:16:
```
$/include/dlwrap.h:131:3: error: static_assert failed due to
requirement '2UL == 1UL' "Arity Error"
static_assert(Requested == Required, "Arity Error");
^ ~~~~~~~~~~~~~~~~~~~~~
$/plugins/cuda/dynamic_cuda/cuda.cpp:23:1: note: in
instantiation of function template specialization 'dlwrap::verboseAssert<2UL, 1UL>' requested
here
DLWRAP_INTERNAL(cuInit, 2);
```
Arity lower than it should be:
Before diff
```
$/plugins/cuda/dynamic_cuda/cuda.cpp:131:10: error: no
matching function for call to 'dlwrap_cuInit'
return dlwrap_cuInit(X);
^~~~~~~~~~~~~
$/plugins/cuda/dynamic_cuda/cuda.cpp:23:1: note: candidate
function not viable: requires 0 arguments, but 1 was provided
DLWRAP_INTERNAL(cuInit, 0);
```
After diff
In file included from $/plugins/cuda/dynamic_cuda/cuda.cpp:16:
```
$/include/dlwrap.h:131:3: error: static_assert failed due to
requirement '0UL == 1UL' "Arity Error"
static_assert(Requested == Required, "Arity Error");
^ ~~~~~~~~~~~~~~~~~~~~~
$/plugins/cuda/dynamic_cuda/cuda.cpp:23:1: note: in
instantiation of function template specialization 'dlwrap::verboseAssert<0UL, 1UL>' requested
here
DLWRAP_INTERNAL(cuInit, 0);
```
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106543
Summary:
Fixes some warning given for uninitialized block counts if the exection mode is
not recognized. This shouldn't happen in practice because the execution mode is
checked when it's read from the device.
This class is instantiated once in rtl.cpp before hsa_init is
called. The hsa_signal_create call therefore fails leaving the pool empty.
This signal pool is a legacy from ATMI where it was constructed after hsa_init.
Moving the state into the rtl.cpp global class disabled the initial populating
of the pool without noticeably changing performance. Just rechecked with a fix
that allocates the signals after hsa_init and that also doesn't noticeably
change performance.
This patch therefore drops the initialisation. Only change from main is to
drop a DEBUG_PRINT statement that would say the pool initial size is zero.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106515
Function internalization can sometimes occur in situations where we want to
keep the call sites intact. This patch adds an option to disable function
internalization and prevents the device runtime from being internalized while
creating the bitcode library.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106438
This patch introduces `__kmpc_is_generic_main_thread_id` which splits the old
comparison into its own runtime function. The purpose of this is so we can fold
this part independently, so when both this and `is_spmd_mode` are folded the
final function will be folded as well.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106437
Qualified kernels can be transformed from generic-mode to SPMD mode using an
optimization in OpenMPOpt. This patch introduces a new execution mode to
indicate kernels that have been transformed from generic-mode to SPMD-mode.
These kernels have SPMD-mode execution, but need generic-mode semantics for
scheduling the blocks and threads. Without this far too few blocks will be
scheduled for a generic region as SPMD mode expects the trip count to be
divided by the number of threads.
Reviewed By: ggeorgakoudis
Differential Revision: https://reviews.llvm.org/D106460
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
The patch exposes the libomptarget runtime function that gets the hardware thread id through the kmpc API. This is to be used in SPMDization for checking the thread id to execute regions by a single thread in a block.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106323
Create a hsa_api.h header that includes the ROCr headers in use
Drop some unused headers and _cplusplus macros
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106455
In `deviceRTLs`, the parallel level is stored in a shared variable of type `uint8_t`.
`__kmpc_parallel_level` currently returns a 16-bit interger. This patch first
changes the return type of the function to `uint8_t`, same as the shared variable,
and then corrects function type which was updated in D105955.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106384
Currently the NPVTX work function is marked volatile. This prevents some
optimizations from using this value.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D106310
D106236 added a new CMake argument for `libomptarget` test, but when user's
input contains white spaces, CMake will add escape char to the final lit command,
which leads to an error. This patch converts the user's input `LIBOMPTARGET_LIT_ARGS`
into a local array, and then passes the array to the function.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D106247
By default, `lit` uses all threads to invoke tests, which can easily cause out
of memory on GPUs because most of OpenMP offloading test usually take about 1GB
GPU memory, but a typical GPU only has 4-8GB memory. This patch introduce a
CMake argument `LIBOMPTARGET_LIT_ARGS` to allow users to control the behavior of
`libomptarget` tests, similar to `LLVM_LIT_ARGS`.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D106236
Currently when we compile the project in debug mode, `-g` will not be added to
compilation flag. The bc files generated in different mode are of different size.
When using GPU debuggers like `cuda-gdb`, it is expected to provide more info
with a debug version of bc lib.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D106229
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
There are two places in current deviceRTLs where it computes parallel level explicitly,
which is basically the functionality of `__kmpc_parallel_level`. Starting from
D105787, we plan to introduce a series of function call folding based on information
that can be deducted during compilation time. Computation of parallel level is
the next target. This patch makes steps for the optimization.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D105955
D105812 introduced a regression where if a PTR_AND_OBJ entry was mapped on the device, then the OBJ was deallocated and then reallocated at a different address, the Shadow Pointer Map would still contain an entry for the PTR but pointing to the old address. This caused test `env/base_ptr_ref_count.c` to fail.
Differential Revision: https://reviews.llvm.org/D105947
Currently, libomptarget will always perform a host-to-device memory transfer in
order to update the device pointer of a PTR_AND_OBJ entry. This is not always
necessary because the device pointer may have been set to the correct pointee
address already, so we can eliminate the redundant memory transfer.
Simplifies control flow to allow store/load forwarding
This change folds two basic blocks into one, leaving a single store to parallelLevel.
This is a step towards spmd kernels with sufficiently aggressive inlining folding
the loads from parallelLevel and thus discarding the nested parallel handling
when it is unused.
Transform:
```
int threadId = GetThreadIdInBlock();
if (threadId == 0) {
parallelLevel[0] = expr;
} else if (GetLaneId() == 0) {
parallelLevel[GetWarpId()] = expr;
}
// =>
if (GetLaneId() == 0) {
parallelLevel[GetWarpId()] = expr;
}
// because
unsigned GetLaneId() { return GetThreadIdInBlock() & (WARPSIZE - 1);}
// so whenever threadId == 0, GetLaneId() is also 0.
```
That replaces a store in two distinct basic blocks with as single store.
A more aggressive follow up is possible if the threads in the warp/wave
race to write the same value to the same address. This is not done as
part of this change.
```
if (GetLaneId() == 0) {
parallelLevel[GetWarpId()] = expr;
}
// =>
parallelLevel[GetWarpId()] = expr;
// because
unsigned GetWarpId() { return GetThreadIdInBlock() / WARPSIZE; }
// so GetWarpId will index the same element for every thread in the warp
// and, because expr is lane-invariant in this case, every lane stores the
// same value to this unique address
```
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D105699
In order to fold calls based on high-level knowledge and control flow
tracking it helps to expose the information as a runtime call. The
logic: `!SPMD && getTID() == getMasterTID()` was used in various places
and is now encapsulated in `__kmpc_is_generic_main_thread`. As part of
this rewrite we replaced eager computation of arguments with on-demand
computation, especially helpful if the calls can be folded and arguments
don't need to be computed consequently.
Differential Revision: https://reviews.llvm.org/D105768
In order to avoid malloc/free, up to NUM_SHARED_VARIABLES_IN_SHARED_MEM
(=64) variables are communicated in dedicated shared memory instead. The
simplification does avoid the need for an "init" and requires "deinit"
only if we ever communicate more than NUM_SHARED_VARIABLES_IN_SHARED_MEM
variables.
Differential Revision: https://reviews.llvm.org/D105767
We had multiple functions to determine the execution mode (SPMD/Generic)
and runtime status (initialized/uninitialized) but that just increased
complexity without a real benefit. Especially with D102307 in mind it
is helpful to reduce the dependence on the `ident_t` flags.
Differential Revision: https://reviews.llvm.org/D105586
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`
We had multiple functions to determine the execution mode (SPMD/Generic)
and runtime status (initialized/uninitialized) but that just increased
complexity without a real benefit. Especially with D102307 in mind it
is helpful to reduce the dependence on the `ident_t` flags.
Differential Revision: https://reviews.llvm.org/D105586
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
This patch is an attempt to do for `targetDataBegin` what D104924 does
for `targetDataEnd`:
* Eliminates a lock/unlock of the data mapping table.
* Clarifies the logic that determines whether a struct member's
host-to-device transfer occurs. The old logic, which checks the
parent struct's reference count, is a leftover from back when we had
a different map interface (as pointed out at
<https://reviews.llvm.org/D104924#2846972>).
Additionally, it eliminates the `DeviceTy::getMapEntryRefCnt`, which
is no longer used after this patch.
While D104924 does not change the computation of `IsLast`, I found I
needed to change the computation of `IsNew` for this patch. As far as
I can tell, the change is correct, and this patch does not cause any
additional `openmp` tests to fail. However, I'm not sure I've thought
of all use cases. Please advise.
Reviewed By: jdoerfert, jhuber6, protze.joachim, tianshilei1992, grokos, RaviNarayanaswamy
Differential Revision: https://reviews.llvm.org/D105121
The patch has the following benefits:
* Eliminates a lock/unlock of the data mapping table.
* Clarifies the logic that determines whether a struct member's
device-to-host transfer occurs. The old logic, which checks the
parent struct's reference count, is a leftover from back when we had
a different map interface (as pointed out at
<https://reviews.llvm.org/D104924#2846972>).
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D104924
If the base is used in a map clause and later we have a memberexpr with
this base, and the member is a pointer, and this pointer is dereferenced
anyhow (subscript, array section, dereference, etc.), such components
should be considered as overlapped, otherwise it may lead to incorrect
size computations, since we try to map a pointee as a part of the whole
struct, which is not true for the pointer members.
Differential Revision: https://reviews.llvm.org/D105562
[libomptarget][nfc] Group environment variables, drop accesses to DeviceInfo global
Folds some duplicates logic into a helper function, passes the new environment
struct into getLaunchVals which no longer reads the DeviceInfo global.
Implemented on top of D105237
Reviewed By: dhruvachak
Differential Revision: https://reviews.llvm.org/D105239
D97883 introduced a compile-time error in the experimental remote offloading
libomptarget plugin, this patch fixes it and resolves a number of
inconsistencies in the plugin as well:
1. Non-functional Asynchronous API
2. Unnecessarily verbose debug printing
3. Misc. code clean ups
This is not intended to make any functional changes to the plugin.
Differential Revision: https://reviews.llvm.org/D105325
`DeviceTy::getOrAllocTgtPtr` just returns a target pointer. In addition,
two bool values (`IsNew` and `IsHostPtr`) are passed by reference to make the
change in the function available in callee.
In this patch, a struct, which contains the target pointer, two flags, and an
iterator to the map table entry corresponding to the queried host pointer, will
be returned. In addition to make the logic clearer regarding the two bool values,
this paves the way for the next patch to fix the data race in `bug49334.cpp` by
attaching an event to the map table entry (and that's why we need the iterator).
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D104382
This reverts commit 2240b41ee4.
A value of 0 for KernDescVal WG_Size implies it is unknown, so it should be
set to the default. The above change was made without this assumption.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D105250
A step towards making this function adequately self contained that it
can be tested easily. No functional change intended here, left variable
names unchanged.
Reviewed By: ronlieb
Differential Revision: https://reviews.llvm.org/D105229
Removes stdarg header, drops uses of iostream, fix some format string errors.
Also changes a C style struct to C++ style to avoid a warning from clang/
Reviewed By: pdhaliwal
Differential Revision: https://reviews.llvm.org/D104923
In our ongoing work, we are using `AbstractAttributor` to deduct execution model
of device functions, and potententially remove unnecessary function calls to
`__kmpc_is_spmd_exec_mode`. In current device runtime, we have mixed use of
`isSPMDMode` and `__kmpc_is_spmd_exec_mode`, but in fact in `__kmpc_is_spmd_exec_mode`
it simply calls `isSPMDMode`. Since all functions starting with `__kmpc` is C
function, which doesn't have things like name mangling. It is more optimization
friendly. In this patch, we simply replaced all calls to `isSPMDMode` with
`__kmpc_is_spmd_exec_mode` to pave the way for the optimization.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D105211
This patch is related to https://reviews.llvm.org/D98832. Based on discussions there, I decided to separate out the teams default as this patch. This change is to increase the number of teams per computation unit so as to provide more wavefronts for hiding latency. This change improves performance for some programs, including 20-50% for some Stream benchmarks.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D99003
When max flat workgroup size is not specified, it is set to the default
workgroup size. This prevents kernel launch with a workgroup size larger
than the default. The fix is to ignore a size of 0 and treat it as
unspecified.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D105073
The logic is almost similar to that of system.cpp with one change that
instead of adding all the memory pools to a device struct it only
keeps a single pool. The existing approach also always allocated memory on
the first HSA pool found for a GPU.
This depends on D104691. The goal of this series of patches is to remove
_atl_machine global. The next patch will drop g_atl_machine entirely.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D104695
[libomptarget][amdgpu] Build openmp for two more targets
The 4800U APU is a gfx902 and the MI100 accelerator is a gfx908.
Both numbers are listed in ROCT topology.c
Reviewed By: jhuber6
Differential Revision: https://reviews.llvm.org/D104922
For example, without this patch:
```
$ cat test.c
int main() {
int x;
#pragma omp target enter data map(alloc: x)
#pragma omp target enter data map(alloc: x)
#pragma omp target enter data map(alloc: x)
#pragma omp target exit data map(delete: x)
;
return 0;
}
$ clang -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda test.c
$ LIBOMPTARGET_DEBUG=1 ./a.out |& grep 'Creating\|Mapping exists\|last'
Libomptarget --> Creating new map entry with HstPtrBegin=0x00007ffddf1eaea8, TgtPtrBegin=0x00000000013bb040, Size=4, RefCount=1, Name=unknown
Libomptarget --> Mapping exists with HstPtrBegin=0x00007ffddf1eaea8, TgtPtrBegin=0x00000000013bb040, Size=4, RefCount=2 (incremented), Name=unknown
Libomptarget --> Mapping exists with HstPtrBegin=0x00007ffddf1eaea8, TgtPtrBegin=0x00000000013bb040, Size=4, RefCount=3 (incremented), Name=unknown
Libomptarget --> Mapping exists with HstPtrBegin=0x00007ffddf1eaea8, TgtPtrBegin=0x00000000013bb040, Size=4, RefCount=2 (decremented)
Libomptarget --> There are 4 bytes allocated at target address 0x00000000013bb040 - is not last
```
`RefCount` is reported as decremented to 2, but it ought to be reset
because of the `delete` map type, and `is not last` is incorrect.
This patch migrates the reset of reference counts from
`DeviceTy::deallocTgtPtr` to `DeviceTy::getTgtPtrBegin`, which then
correctly reports the reset. Based on the `IsLast` result from
`DeviceTy::getTgtPtrBegin`, `targetDataEnd` then correctly reports `is
last` for any deletion. `DeviceTy::deallocTgtPtr` is responsible only
for the final reference count decrement and mapping removal.
An obscure side effect of this patch is that a `delete` map type when
the reference count is infinite yields `DelEntry=IsLast=false` in
`targetDataEnd` and so no longer results in a
`DeviceTy::deallocTgtPtr` call. Without this patch, that call is a
no-op anyway besides some unnecessary locking and mapping table
lookups.
Reviewed By: grokos
Differential Revision: https://reviews.llvm.org/D104560
For example, without this patch:
```
$ cat test.c
int main() {
int x;
#pragma omp target enter data map(alloc: x)
#pragma omp target exit data map(release: x)
;
return 0;
}
$ clang -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda test.c
$ LIBOMPTARGET_DEBUG=1 ./a.out |& grep 'Creating\|Mapping exists'
Libomptarget --> Creating new map entry with HstPtrBegin=0x00007ffcace8e448, TgtPtrBegin=0x00007f12ef600000, Size=4, Name=unknown
Libomptarget --> Mapping exists with HstPtrBegin=0x00007ffcace8e448, TgtPtrBegin=0x00007f12ef600000, Size=4, updated RefCount=1
```
There are two problems in this example:
* `RefCount` is not reported when a mapping is created, but it might
be 1 or infinite. In this case, because it's created by `omp target
enter data`, it's 1. Seeing that would make later `RefCount`
messages easier to understand.
* `RefCount` is still 1 at the `omp target exit data`, but it's
reported as `updated`. The reason it's still 1 is that, upon
deletions, the reference count is generally not updated in
`DeviceTy::getTgtPtrBegin`, where the report is produced. Instead,
it's zeroed later in `DeviceTy::deallocTgtPtr`, where it's actually
removed from the mapping table.
This patch makes the following changes:
* Report the reference count when creating a mapping.
* Where an existing mapping is reported, always report a reference
count action:
* `update suppressed` when `UpdateRefCount=false`
* `incremented`
* `decremented`
* `deferred final decrement`, which replaces the misleading
`updated` in the above example
* Add comments to `DeviceTy::getTgtPtrBegin` to explain why it does
not zero the reference count. (Please advise if these comments miss
the point.)
* For unified shared memory, don't report confusing messages like
`RefCount=` or `RefCount= updated` given that reference counts are
irrelevant in this case. Instead, just report `for unified shared
memory`.
* Use `INFO` not `DP` consistently for `Mapping exists` messages.
* Fix device table dumps to print `INF` instead of `-1` for an
infinite reference count.
Reviewed By: jhuber6, grokos
Differential Revision: https://reviews.llvm.org/D104559
The OpenMP 5.1 standard defines the environment variable
`OMP_TEAMS_THREAD_LIMIT` to limit the number of threads that will be run in a
single block. This patch adds support for this into the AMDGPU and CUDA
plugins.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D103923
Currently the runtime implementation of `__kmpc_alloc_shared` is extremely slow because it allocated memory for each thread individually. This patch adds a small buffer for the threads to share data and will greatly improve performance for builds where all globalization could not be optimized out. If the shared buffer is full, then memory will not only be allocated per-warp rather than per-thread.
Depends on D97680
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D104666
Summary:
This patch introduces the new globalization runtime to be used by D97680. These
runtime calls will replace the __kmpc_data_sharing_push_stack and
__kmpc_data_sharing_pop_stack functions.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D102532
There does not seem to be any use of these functions. They just
put the value to a local which is never used again.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D104512
`bug49334.cpp` cannot detect data race in `libomptarget` efficiently. It
is reported that with `N = 256` and `BS = 16`, the data race can be reproduced
more steadily. The next coming pathces will fix it so this patch is expected to
fail now.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D104552
This change-set removes libelf usage from elf_common part of the plugins.
libelf is still used in x86_64 generic plugin code and in some plugins
(e.g. amdgpu) - these will have to be cleaned up in separate checkins.
Differential Revision: https://reviews.llvm.org/D103545
This patch includes some changes which deletes the code accessing
g_atl_machine global. Some accesses related to memory_pools are
still remaining.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D103813
This patch adds an information flag that indicated when data is being copied to
and from the device. This will be helpful for finding redundant or unnecessary
data transfers in applications.
Reviewed By: jdoerfert, grokos
Differential Revision: https://reviews.llvm.org/D103927
This global struct used to hold various flags for monitoring the
initialization of hsa.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D103795
Previous logic was to always use the first kernarg pool found to allocate
kernel args. This patch changes this to use only the kernarg pool which
has non-zero size. This logic is also reworked to not use any globals.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D103600
Turns out the only purpose of this class was verify if device ID
was in range or not which could be done easily by using g_atl_machine.
Still getting rid of g_atl_machine is pending which would be done in
a later patch.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D103443
This struct was used to specify the device on which memory was
being allocated/free in atmi_malloc/free. It has now been replaced
with int DeviceId.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D103239
Refactor suggested in D103037 to help avoid similar copy-paste errors.
Change is mechanical. Some parts of this would be more robust with unsigned.
Reviewed By: dhruvachak
Differential Revision: https://reviews.llvm.org/D103090
Suggested in D103059. Use a single lookup instead of two, more const, less mutation.
Reviewed By: dhruvachak
Differential Revision: https://reviews.llvm.org/D103093
ATMI_STATUS_UNKNOWN was unused, deleted references to it.
Replaced ATMI_STATUS_{SUCCESS,ERROR} with HSA_STATUS_{SUCCESS,ERROR}
Replaced atmi_status_t with hsa_status_t
Otherwise no change. In particular, conversions between atmi_status_t and
hsa_status_t will now be conversions between hsa_status_t and itself.
Reviewed By: pdhaliwal
Differential Revision: https://reviews.llvm.org/D103115
This patch drops g_atmi_initialized and inlines the Initialize &
Finalize methods from Runtime class.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D102847
Two globals KernelInfoTable & SymbolInfoTable are moved
into RTLDeviceInfoTy class.
This builds on the top of D102691.
[2/2]
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D102692
[libomptarget][nfc] Move hostcall required test to rtl
Remove a global, fix minor race. First of N patches to bring up hostcall.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D103058
Fix the case where NumTeams was set incorrectly instead of NumThreads
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D103037
KernelNameMap contains entries like "key.kd" => key which clearly
could be replaced by simple logic of removing suffix from the key.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D102691
The check for the TO flag when processing firstprivates is missing. As a result,
sometimes the device copy of a firstprivate never gets initialized. Currectly we
try to force lambda structs to be allocated immediately by marking them as a
non-firstprivate, so that PrivateArgumentManagerTy::addArg allocates memory for
them immediately. However, calling addArg with IsFirstPrivate=false makes the
function skip initializing the device copy. Whether an argument is firstprivate
and whether we need to allocate memory immediately are not synonyms, so this
patch introduces one more control variable for immediate allocation and sets it
apart from initialization.
Differential Revision: https://reviews.llvm.org/D102890
[libomptarget][amdgpu] Mark alloc, free weak to facilitate local experimentation
There are a lot of different ways we might implement the devicertl local alloc
and free functions. Via host, local buffers (stack or arena), specialising per
kernel etc. It is not yet clear what the right design is. This change makes the
alloc and free functions weak, so one can override them from local tests while
comparing options.
Not strictly necessary, as a comparable patch can be applied locally each time,
but would be convenient for out of tree dev. Plan would be to drop the weak
attribute at the same time as introducing a working allocator to trunk.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D102499
[libomptarget] Improve dlwrap compile time error diagnostic
The dlwrap interface takes an explict arity, e.g. DLWRAP(cuAlloc, 2);
This probably can't be eliminated as it controls the argument list of an
external symbol, not an inline header function. If the arity given is too
big, the error from clang referring to the line is in the middle of
implementation details.
/usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10/tuple:1277:7: error: static_assert failed
due to requirement '0UL < tuple_size<std::tuple<>>::value' "tuple index is in range"
static_assert(__i < tuple_size<tuple<>>::value,
^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10/tuple:1260:7: ...
/usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10/tuple:1260:7: ...
/home/amd/llvm-project/openmp/libomptarget/include/dlwrap.h:93:27 ...
/home/amd/llvm-project/openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.cpp:34:1: note: in
instantiation of template class 'dlwrap::trait<cudaError_enum (*)(unsigned long *, unsigned
long)>::arg<2>' requested here
DLWRAP(cuMemAlloc, 3);
^
/home/amd/llvm-project/openmp/libomptarget/include/dlwrap.h:51:31: ...
/home/amd/llvm-project/openmp/libomptarget/include/dlwrap.h:166:3: ...
/home/amd/llvm-project/openmp/libomptarget/include/dlwrap.h:133:3: ...
/home/amd/llvm-project/openmp/libomptarget/include/dlwrap.h:186:37: ...
If the arity is too small, the diagnostic is better:
cuda/dynamic_cuda/cuda.cpp:34:1: error: too few
arguments to function call, expected 2, have 1
DLWRAP(cuMemAlloc, 1);
This patch changes the diagnostic to:
cuda/dynamic_cuda/cuda.cpp:34:1: error:
static_assert failed due to requirement '1 == trait<cudaError_enum (*)(unsigned long *, unsigned
long)>::nargs' "Arity Error"
DLWRAP(cuMemAlloc, 1);
or
cuda/dynamic_cuda/cuda.cpp:34:1: error:
static_assert failed due to requirement '3 == trait<cudaError_enum (*)(unsigned long *, unsigned
long)>::nargs' "Arity Error"
DLWRAP(cuMemAlloc, 3);
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D102858
[libomptarget][amdgpu] Remove majority of fatal errors
Replaces most calls to exit() with returning an error to the library entry
point. Minor changes to error handling for clear bugs, remove some dead code.
Each exit() call site replaced is either in a library entry point or a
function that already returns error codes on some paths. The existing handling
is not well tested but replacing exit() with a fallback path should be a strict
improvement.
Remaining two early exit points are an abort() from a callback and exit() from
within msgpack. Fixes for those are less obvious and left for a later patch.
Reviewed By: pdhaliwal
Differential Revision: https://reviews.llvm.org/D102346
[libomptarget] Disable test bug49334 on amdgpu
Hangs on amdgpu, do not know why. Disable to unblock build.
Reviewed By: ye-luo
Differential Revision: https://reviews.llvm.org/D102017
This patch moves g_executables to private member of Runtime class
and is renamed to HSAExecutables following LLVM naming convention.
This movement required making Runtime::Initialize and Runtime::Finalize
non-static. Verified the correctness of this change by running
libomptarget tests on gfx906.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D102600
This initial patch removes some unused variables from global namespace.
There will more incoming patches for moving global variables to classes
or static members.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D102598
[libomptarget][amdgpu] Fix truncation error for partial wavefront
The partial barrier implementation involves one wavefront resetting and N-1
waiting. This change future proofs against launching with a number of threads
that is not a multiple of the wavefront size.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D102407
[libomptarget][amdgpu] Convert an assert to print and offload_fail
The kernel launched is supposed to be present in the binary, but a not yet
diagnosed bug means it is missing for some of the qmcpack test cases. Changing
from assert to print and offload_fail should help diagnose that and similar bugs.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D102378
Add a `REQUIRES: unified_shared_memory` option to tests that use `#pragma omp requires unified_shared_memory`.
For CUDA, the feature tag is derived from LIBOMPTARGET_DEP_CUDA_ARCH which itself is derived using [[ https://cmake.org/cmake/help/latest/module/FindCUDA.html#commands | cuda_select_nvcc_arch_flags ]]. The latter determines which compute capability the GPU in the system supports. To ensure that this is the CUDA arch being used, we could also set the `-Xopenmp-target -march=` flag.
In the absence of an NVIDIA GPU, LIBOMPTARGET_DEP_CUDA_ARCH will be 35. That is, in that case we are assuming unified_shared_memory is not available. CUDA plugin testing could be disabled entirely in this case, but this currently depends on `LIBOMPTARGET_CAN_LINK_LIBCUDA OR LIBOMPTARGET_FORCE_DLOPEN_LIBCUDA`, not on whether the hardware is actually available.
For all other targets, nothing changes and we are assuming unified shared memory is available. This might need refinement if not the case.
This tries to fix the [[ http://meinersbur.de:8011/#/builders/143 | OpenMP Offloading Buildbot ]] that, although brand-new, only has a Pascal-generation (sm_61) GPU installed. Hence, tests that require unified shared memory are currently failing. I wish I had known in advance.
Reviewed By: protze.joachim, tianshilei1992
Differential Revision: https://reviews.llvm.org/D101498
[libomptarget][amdgpu][nfc] Expand errorcheck macros
These macros expand to continue, which is confusing, or exit,
which is incompatible with continuing execution on offloading fail.
Expanding the macros in place makes the code look untidy but the
control flow obvious and amenable to improving. In particular, exit
becomes easier to eliminate.
Reviewed By: pdhaliwal
Differential Revision: https://reviews.llvm.org/D102230
Pushpinder Singh