GFX 9.4.x has better performance for CPU-GPU copies when using
engines in reverse order from other devices.
Change-Id: I1eaebf0e837bb7f44712f40d5115df618f6a73d7
If the KFD doesn't support targeting SDMA engines, ensure that ROCr
selects the correct downstream queue type by using an invalid engine.
Change-Id: Ia6848126f67f3d35ab37248633e8e0e6e2d77fff
- Replaces logic relying on clock64() instead to
use an atomic counter that the last wg in grid
and multi-grid groups wait on while the the
non-last work groups increment the counter
to "done"
- Implements a new logic to verify the multi-grid
sync()
Change-Id: I7780d8124e5f144b124e5d191d0f412483a3b565
To support fully-static library ROCm builds, ensure that all global
symbols are prefixed with something meaningful to avoid collisions with
other libraries
A script was made using" objdump -C -t" to get a list of symbols,
then checking if the global symbols have a meaningful prefix (for thunk:
hsakmt or kmt in various cases)
Change-Id: Ifd353f64a3344eb60d1f6c4e041aa20967b38a59
Signed-off-by: Kent Russell <kent.russell@amd.com>
- Use HSA_ALLOCATE_QUEUE_DEV_MEM=1 to create AQL queue in device
memory.
- Before writing AQL packet header to the queue use an SFENCE to ensure
that there is no reodering of the writes over PCIE
Change-Id: I5eacdc35108c4a1e245c75ae349b7495451aa60d
Fix for some places where the ISA buffers are not declared as
executable. Previous code in Thunk was blindly setting exec bit on all
memory allocations so this issue was masked.
Change-Id: Ic7a1169c69fb85ff9e8ea7bcc49a1845b37c08ff
* PC sampling: online partial PC sampling decoding
PC sampling service decodes a PC sample partially
by replacing the PC with an id of the loaded code object instance
containing PC and the offset of the PC within that code object instance.
* PC sampling: marker records removed
* PC sampling parser: minor doc update in mock
* PC sampling: introducing rocprofiler_pc_t
* NULL value of the code object id introduced.
* Clarifying documenation related to PC offset.
* PC offset documentation improvement
* PC sampling parser benchmark: Reducing the number of samples to recreate half of performance.
[ROCm/rocprofiler-sdk commit: 93e82663d9]
* PC sampling: online partial PC sampling decoding
PC sampling service decodes a PC sample partially
by replacing the PC with an id of the loaded code object instance
containing PC and the offset of the PC within that code object instance.
* PC sampling: marker records removed
* PC sampling parser: minor doc update in mock
* PC sampling: introducing rocprofiler_pc_t
* NULL value of the code object id introduced.
* Clarifying documenation related to PC offset.
* PC offset documentation improvement
* PC sampling parser benchmark: Reducing the number of samples to recreate half of performance.
The function can return NULL if it fails to create the backend, so check
for NULL before using it.
Change-Id: I4d6501bffd6dd0fc0d0f2224720f7d6dca1646f3
Signed-off-by: Kent Russell <kent.russell@amd.com>
This commit disables samples that directly call hipcc when
building in a static environment. In its current state, hipcc
requires all static libraries to be explicitly provided
in order to succesfully link the sample binary.
Change-Id: I74c94a57867bc1fe0e6134ddd633b465bd6c0f83
This reverts commit 9f0f7741de.
For APU, the PCIe atomic is supported by default. However, the PCIe
atomic feature needs to checked for dGPU. The kfd driver has already
set PCIe atomic support for APUs, so this patch can be reverted.
Change-Id: I131d5b8e095c1104e1695e7cf8b1ed178bccddde
Signed-off-by: Shane Xiao <shane.xiao@amd.com>
rocm-clang-ocl has been deprected from ROCm 6.1 onwards. Uprading ROCm from 6.0 to latest release will result in stale files of rocm-clang-ocl.
rocm-core will replaces/obsoletes rocm-clang-ocl package. This will ensure clean removal of the deprecated package
Change-Id: Id5c16492273d8a0dcbbd5ac797daaef96c058f5b
On Fedora, rocm-smi is a standard package and is installed to /usr/bin
So when run_kfdtest.sh is run this error is produced
find: ‘/opt/rocm*’: No such file or directory
First redirect stderr to dev/null on the original search.
Then fall back to either looking for rocm-smi in BIN_DIR or
look for it in the PATH.
Change-Id: I389ed0b9a4a4507263c9eb19894b25326c9a4222
Signed-off-by: Tom Rix <Tom.Rix@amd.com>
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
Using "PROGRAMS" and "FILES" without specifying permissions will
automatically select the right permissions.
PROGRAMS is used for executables, FILES is used for data files
Change-Id: I0fb6eff257a8f936848bd648cf877da6dc0b6906
Fix segfault on p2p copies when 2 agents cannot access each other's
memory (usually because the PCI BAR's are out of range). The
AcquireAsyncCopyAccess function should return NULL in that case, so that
the test can be skipped.
Change-Id: If018f3609dd21a01c56eaec94de3bca52c385c4d
Jonathan Kim
Sourabh Betigeri
kjayapra-amd
Vladimir Indic
Kent Russell
Chris Freehill
Saleel Kudchadker
David Galiffi
David Yat Sin
Julia Jiang
victzhan
Ammar ELWazir
Ioannis Assiouras
Shane Xiao
dependabot[bot]
Nirmal Unnikrishnan
AravindanC
Ranjith Ramakrishnan
Vladana Stojiljkovic
Rahul Manocha
Jeremy Newton
Xuanteng Huang
Jaydeep Patel
Tom Rix
Marko Arandjelovic
Jimbo Xie
Joseph Greathouse
Giovanni LB