caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[OpenMP] Add environment variables to change stack / heap size in the CUDA plugin

Browse Source 
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
This commit is contained in:
Joseph Huber
2021-07-22 21:19:14 -04:00
committed by Huber, Joseph
parent 4a76bd0e31
commit 3817ba13ae
2 changed files with 40 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
openmp/docs/design/Runtimes.rst
View File

@@ -30,6 +30,8 @@ variables is defined below.
* ``LIBOMPTARGET_PROFILE=<Filename>``
* ``LIBOMPTARGET_MEMORY_MANAGER_THRESHOLD=<Num>``
* ``LIBOMPTARGET_INFO=<Num>``
* ``LIBOMPTARGET_HEAP_SIZE=<Num>``
* ``LIBOMPTARGET_STACK_SIZE=<Num>``
LIBOMPTARGET_DEBUG
""""""""""""""""""
@@ -321,6 +323,21 @@ default. The solution is to add an explicit map clause in the target region.
return sum;
}
LIBOMPTARGET_STACK_SIZE
"""""""""""""""""""""""
This environment variable sets the stack size in bytes for the CUDA plugin. This
can be used to increase or decrease the standard amount of memory reserved for
each thread's stack.
LIBOMPTARGET_HEAP_SIZE
"""""""""""""""""""""""
This environment variable sets the amount of memory in bytes that can be
allocated using ``malloc`` and ``free`` for the CUDA plugin. This is necessary
for some applications that allocate too much memory either through the user or
globalization.
.. toctree::
:hidden:
:maxdepth: 1

23
openmp/libomptarget/plugins/cuda/src/rtl.cpp
View File

@@ -642,11 +642,34 @@ public:
DeviceData[DeviceId].BlocksPerGrid = EnvTeamLimit;
}
size_t StackLimit;
size_t HeapLimit;
if (const char *EnvStr = getenv("LIBOMPTARGET_STACK_SIZE")) {
StackLimit = std::stol(EnvStr);
if (cuCtxSetLimit(CU_LIMIT_STACK_SIZE, StackLimit) != CUDA_SUCCESS)
return OFFLOAD_FAIL;
} else {
if (cuCtxGetLimit(&StackLimit, CU_LIMIT_STACK_SIZE) != CUDA_SUCCESS)
return OFFLOAD_FAIL;
}
if (const char *EnvStr = getenv("LIBOMPTARGET_HEAP_SIZE")) {
HeapLimit = std::stol(EnvStr);
if (cuCtxSetLimit(CU_LIMIT_MALLOC_HEAP_SIZE, HeapLimit) != CUDA_SUCCESS)
return OFFLOAD_FAIL;
} else {
if (cuCtxGetLimit(&HeapLimit, CU_LIMIT_MALLOC_HEAP_SIZE) != CUDA_SUCCESS)
return OFFLOAD_FAIL;
}
INFO(OMP_INFOTYPE_PLUGIN_KERNEL, DeviceId,
"Device supports up to %d CUDA blocks and %d threads with a "
"warp size of %d\n",
DeviceData[DeviceId].BlocksPerGrid,
DeviceData[DeviceId].ThreadsPerBlock, DeviceData[DeviceId].WarpSize);
INFO(OMP_INFOTYPE_PLUGIN_KERNEL, DeviceId,
"Device heap size is %d Bytes, device stack size is %d Bytes per "
"thread\n",
(int)HeapLimit, (int)StackLimit);
// Set default number of teams
if (EnvNumTeams > 0) {