HDP will now be used for coarse grain kernarg so needs to be
reported without consideration of fine grain vram over pcie.
Change-Id: I648167299faa583876a3d8685c3b3c4d8d31ebf9
Setting to 1 prevents the scratch handler from reducing peak occupancy.
Scratch allocations that would normally reduce peak occupancy will
instead fail.
Diagnostic for TF and PyTorch.
Change-Id: I2d7ea47077eb5cf708251c8aa3fd183ad4261be0
scratch_used_large_ was uninitialized leading to the observed hang.
DynamicScratchHandler would wait for a large scratch release despite no
large scratch having yet been allocated. Fixes .
The patch also removes a potential race between AddScratchNotifier and
ReleaseQueueScratch. The race condition does not exist today since both
scratch alloc and release run on the same thread. The changes will
prevent this potential race from manifesting if the async event handler
is ever updated to use multiple threads.
Also enhances scratch occupancy reduction reporting. Reporting now
prints the initial request size as well as the allocated size and the
effect on occupancy this has. Occupancy is computed in terms of the
requesting dispatch grid size so may be >100%.
Change-Id: I0fc5ee01467ff4c29bdd25d545177c97862c3bd9
Ensures that all CPU agents will have a pool handle to allocate
system memory. These pools will have no numa binding since the
node their owning Agent represents has no installed memory.
Change-Id: I9f72b455d633646839753c6719ff7f6a4c41f7c4
- This new path is required when libhsaruntime.so is referred
from the top level ROCm lib directory.
- Once ROCm stack lib/lib64 structure is flatten, RUNPATH in all
the libraries needs to be updated.
Change-Id: I369131ce93e14958ec57a54701671f2bfd8d522a
Signed-off-by: Pruthvi Madugundu <pruthvi.madugundu@amd.com>
Move the shader code before the test case code so that all test case
code is consecutive.
Rectify the print messages and avoid calling GetSysMemSize() repeatedly.
Change-Id: I1c4aa5552de4d74163717fe66ad9759fb09e1316
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
The original test takes forever to run on emulators because emulators
are much slower than Asic. So intelligently detect the emulator scenarios
and reduce the run time by slashing the iteration times.
Change-Id: I087f43c04c2b23b5ab2ecaad07533b767c337e94
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
Use an "if 0" to remove the code, as it is not working as expected. The
test is supposed to test mapping from a foreign device (like a NIC), so
it uses a separate GPU to map, but this mapping can be evicted and thus
the test can fail unexpectedly. Remove the code until the test can be
reworked
Change-Id: Ie4a15c2a018bbd8e931b06b6700d10b3be86e410
This will allow it to be installed with the ROCm suite,
and centralize things a little bit more
Also update run_kfdtest.sh to reflect the changes
Lastly, remove "die" reference as compute_utils.sh
may not be packaged with KFDTest
Change-Id: I4c30cd29979192496419e71e3685937d7417f739
If child process explicitly calls hsaKmtReleaseSystemProperties(), it
fails. Allow child process to release and acquire system properties.
Change-Id: I649a4600212711b2ad4474f605f3ca39a4003d03
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Tests run in the order in which they are linked. Currently that order
is non-deterministic. Listing source files in the Makefile explicitly
makes the order deterministic.
The order chosen runs basic tests before more advanced tests before
stress tests.
Change-Id: I5bc032bcd589f92a51db36acb518bb4d8ef778d3
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
IPC memory was previously returned as HSA_POINTER_ALLOCATED and
had garbage in the node_id field. Due to ROCR_VISIBLE_DEVICES
we need to be able to distinguish between imported memory and
regular memory because imported memory may not be owned by an
agent that is visible in the process. Differentiating these flags
allows the users to expect null agent for the owning agent.
Fixes
Change-Id: Ide3489cec1ee2072dc9697fa5cb71ddb17999d14
The name changed and thus will end up throwing errors if an upgrade is
done, compared to a plain install
Change-Id: Ibc7876a66a414034a00f924cdd750e6a08d6c9cc
From Thunk spec, flag NoSubstitute = 0, if specific memory type is not
available on node, allocation may fall back to other memory that can
replace it on that node. mbind return failure if no memory available on
the specific node, we should ignore the mbind failure for this case.
Change-Id: I651a1bedf1852330604e56965cc17862403ebf87
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
when variable interval become zero, the loop of while (timeConsumption >= interval)
will be a infinite loop. The fix is to avoid that case.
Change-Id: I8fd07296925300bace5ab7d3da86482b6d8b0d03
Signed-off-by: Eric Huang <JinhuiEric.Huang@amd.com>
Attribute optimize(0) doesn't appear to be helpful helpful. This
prevents optimization in the function but not at call sites to the
function. The function may still be inlined since it has no side
effect (in some cases that we currently don't support).
Having a side effect prevents a call site optimization that allows
removal of a noinline function call with no side effect. Call site
optimization should only happen (in GCC at least) when using whole
program optimization so this may be stronger than we strictly need.
Also added _amdgpu_r_debug to the exported symbol list (global) and
switched to the standard macro for an exported symbol (HSA_API).
Without being in the global list the debugger will not find this
symbol if the binary has been stripped.
Change-Id: Ieb00175ccc55fda4491deee44711cd55b3f24aeb
Check for errno == EBADF in kmtIoctl to detect misuse of the kfd_fd
in a forked child process.
Detect being in a forked child process pro-actively by implementing
a pthread_atfork callback.
Make sure all mutexes get reinitialized in the child process to avoid
deadlocks.
Check for being in a forked child process in CHECK_KFD_OPENED so that
all hsaKmt functions will return the appropriate status
HSAKMT_STATUS_KERNEL_IO_CHANNEL_NOT_OPENED.
Update InvalidKFDHandleTest to expect that error code.
Change-Id: I0238e5fba344dcaa454e97a35db2e2dcc8d1f607
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Adding patch number based on ROCM build/release to have unique
file name for libraries across multiple versions of ROCM.
Signed-off-by: Pruthvi Madugundu <pruthvi.madugundu@amd.com>
Change-Id: I58d665b0e7d577b5bd7a6000d1202a0242672727
After XGMI SDMA queues were separated from regular SDMA queues, they
were not covered in the current tests. Add tests for them now.
Change-Id: I036e3ca5d583ab7f022a9dc6cda3ef867f4773a0
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
On gfx1012, allocating 1/4 of the system memory on a 32G RAM machine
could fail, resulting in this test to fail. Limit the maximum buffer
to allocate to be smaller than 3G to accommodate this situation.
Change-Id: I38b0a0f7da1f0b9ca851e04d2d0a51767858c801
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
CMake 3.6.0 and older don't have CPACK_RPM_FILE_NAME support, so make
the packaging simpler by checking the OS. If it's Ubuntu, call the dev
package hsakmt-roct-dev, otherwise call it hsakmt-roct-devel, since
Ubuntu is the only officially-ROCm-supported deb-based OS, and
deb-based dev packages end with -dev (while rpm-based use -devel)
Change-Id: If75324f82f507c3b312bb6176c06643d521ccb68
NumCpQueues and NumSdmaQueuesPerEngine should be got by kfd driver not hardcode.
So add two data fields in HsaNodeProperties then thunk is able to get it from sysfs
that exposed by kfd.
v2: change NumCpQueues/NumSdmaQueuesPerEngine to one byte.
v3: merge two commits as one to avoid ABI update two times.
Change-Id: Ie386e4685f13493e22db6e207a399db6a4c5b9dc
Signed-off-by: Huang Rui <ray.huang@amd.com>
The shader supports multiple work items. It also eliminates one input buffer.
Change-Id: If0596b306065980b74fb92613e95610defc00164
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
Lack of cache controls only allow operating SDMA at
agent scope. All copy APIs are defined at system scope so may
result in data errors.
Change-Id: I9cd10007defddcbf8feb14a2e3daa1ba17c0489f
The initial baseline measurements are proving inconsistent, which
results in the test failing more often with different variant rates
Change-Id: I1f4e04bf7d615cf39de9605bd5141a997b22cdfc
Instead of letting CPack generate the filename according to
CPACK_PACKAGE_NAME}-${CPACK_PACKAGE_VERSION}-${CPACK_SYSTEM_NAME},
set the package name ourselves. Also take the change to fix the name of
hsakmt-dev to hsakmt-devel for RPM, since that's their convention
Previous packaging format:
hsakmt-roct-1.0.9-229-g2144854-Linux.deb
hsakmt-roct-1.0.9-229-g2144854-Linux.rpm
hsakmt-roct-dev-1.0.9-229-g2144854-Linux.deb
hsakmt-roct-dev-1.0.9-229-g2144854-Linux.rpm
New format:
hsakmt-roct-1.0.9-292-gc66f8cf.x86_64.deb
hsakmt-roct-1.0.9-292-gc66f8cf.x86_64.rpm
hsakmt-roct-dev-1.0.9-292-gc66f8cf.x86_64.deb
hsakmt-roct-devel-1.0.9-292-gc66f8cf.x86_64.rpm
Change-Id: I4fc4e0fd2eafd25669c1cfffb39860e25a0b645c
Signed-off-by: Kent Russell <kent.russell@amd.com>
The old names are not accurate enough and we rename them according to
their corresponding fault types.
Change-Id: Icf4d52ba0ab9d49af5d912a0feb82665b1e8d344
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
The InvalidPPR* tests are only useful for gfx801 right now, on which
they won't trigger exceptions. So they are not relevent in the
KFDExceptionTest category. In addition, given AccessPPRMem already tests
the PPR memory functionality, we can just delete those two tests.
Change-Id: Id5c6e23c4c0ce47a4f04e9e1f0fa9083e0a9d0e0
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
This puts all CP and SDMA queues in a single test, which is
currently missing.
Change-Id: I98bf58df1be65fe9daf6311c016a48569a8ab674
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
To keep normal performance, we have to use symmetrical cu masks
based on shader engines. So change baseline from 1 cu to 1 cu
per SE.
Change-Id: I9e83a87fb670bb406f7983714fa0d8ab673609eb
Signed-off-by: Eric Huang <JinhuiEric.Huang@amd.com>
Queues should transition to ref counting for all queues eventually.
That cleanup will be part of shared queue pooling support.
Change-Id: I217ff5d573156678b9559da6fb81baa8cd31c617
Sean Keely
Ramesh Errabolu
Saleel Kudchadker
Chris Freehill
Pruthvi Madugundu
Yong Zhao
Kent Russell
Harish Kasiviswanathan
Felix Kuehling
Philip Yang
Eric Huang
Freddy Paul
Laurent Morichetti
Srinivasan Subramanian
Qingchuan Shi
Oak Zeng
Huang Rui
Konstantin Zhuravlyov