Replace direct hsakmt API calls with calls through the driver abstraction layer
in queue management related functions. This includes:
- CreateQueue/DestroyQueue operations
- Queue update and GWS allocation
- CU masking configuration
Also update the corresponding error status types from HSAKMT_STATUS to
hsa_status_t and adjust error handling accordingly.
Signed-off-by: Honglei Huang <Honglei1.Huang@amd.com>
Replace direct libhsakmt calls with driver interface methods
in GpuAgent initialization:
- Replace hsaKmtGetNodeMemoryProperties with driver().GetMemoryProperties
- Replace hsaKmtGetNodeCacheProperties with driver().GetCacheProperties
Signed-off-by: Honglei Huang <Honglei1.Huang@amd.com>
The agent properties variable `agent_props` was declared but never used
in the `InitScratchSRD()` function. Which casued compile warning:
runtime/core/runtime/amd_aql_queue.cpp:1880:15: warning:
unused variable ‘agent_props’ [-Wunused-variable]
1880 | const auto& agent_props = agent_->properties();
No functional changes, purely a code cleanup commit.
Support has been added to query the following
HSA_AMD_INFO_GET_CLOCK_COUNTERS agent info exposed through the hsa api
in rocr, rather than the user having to make a direct IOCTL call
through the kernel driver.
Signed-off-by: Sunday Clement <Sunday.Clement@amd.com>
Extend hsa_amd_vmem_address_reserve/hsa_amd_vmem_address_reserve_align
to support HSA_AMD_VMEM_ADDRESS_NO_REGISTER flag. This allocation can be
used to reserve virtual address ranges that can later be used by
hsa_amd_svm_attributes_set for SVM based memory allocations.
scratch_cache.h includes amd_gpu_agent.h which then again includes
scratch_cache.h, this has now been fixed removing the unecessary
header include.
Signed-off-by: Sunday Clement <Sunday.Clement@amd.com>
scratch_backing_memory_byte_size was originally removed, and then put
back in 02b38d0614. This was because it
was used by rocgdb. rocgdb code has been updated to not use this field.
Bumped _amdgpu_r_debug for the ABI change.
The original version of hsa_amd_portable_export_dmabuf() did not
consider the conditions under which a dmabuf could be shared.
In the new version (hsa_amd_portable_export_dmabuf_v2()), the caller
can specify the flag HSA_AMD_DMABUF_MAPPING_TYPE_PCIE, which means they
want to share the dmabuf over PCIe. In that case, the new code will check
that if it is a PCIe GPU and it is not in a XGMI Hive then if
large-BAR is not supported, we will return an error.
The original version of hsa_amd_portable_export_dmabuf() did not
consider the conditions under which a dmabuf could be shared.
In the new version (hsa_amd_portable_export_dmabuf_v2()), the caller
can specify the flag HSA_AMD_DMABUF_MAPPING_TYPE_PCIE, which means they
want to share the dmabuf over PCIe. In that case, the new code will check
that if it is a PCIe GPU and it is not in a XGMI Hive then if
large-BAR is not supported, we will return an error.
Its safer to have the integer literal explicitly be an unsigned long
in this expression as that's what the type of the errorCode variable
resolves to, preventing any overflow errors.
Signed-off-by: Sunday Clement <Sunday.Clement@amd.com>
ehdr->e_shentshize and ehdr->e_shnum are both 16-bit unsigned integers
and so their types get implicitly promoted to signed int automatically
during the multiplication, they must be explicitly cast into a larger
unsigned type, otherwise if the signed product is large enough the
value is sign extended resulting in incorrect values.
Signed-off-by: Sunday Clement <Sunday.Clement@amd.com>
Ensure file descriptor 'in' is properly closed in error cases
when calling _lseek() during readFrom() operations.
Fix potential resource leak when errors occur during file operations.
Signed-off-by: Alysa Liu <Alysa.Liu@amd.com>
Because eventDescrp->mutex is a non-recursive lock attempting to
acquire the lock with pthread_mutex_lock can cause the system to hang
indefinitely if the lock was already previously aquired with the
preceeding call to pthread_mutex_trylock.
Signed-off-by: Sunday Clement <Sunday.Clement@amd.com>
Refactor variable assignments to use std::move() where appropriate.
Updat function headers to accept parameters by const& where appropriate.
Signed-off-by: Alysa Liu <Alysa.Liu@amd.com>
Changed variable assignments to use std::move() where appropriate.
Changed function headers to pass string arguments by reference where appropriate.
Signed-off-by: Alysa Liu <Alysa.Liu@amd.com>
ROCr initially had a bug where memory allocations that were not 4K
aligned were internally 4K aligned but ROCr would not keep track
of user-requested size. This would cause some pointer_info queries
to fail, but HIP was already aligning the buffer sizes for IPC
requests. For backward compatibility accross 2 minor versions,
we allowed IPC look-ups to be both aligned and un-aligned.
Removing this check as this 4 minor versions have been released
since then.
disable KFD RAS test case as the tests cause GPU reset
which affects the active kfdtest, the tests can only be
run successfully as separate processes
Signed-off-by: Apurv Mishra <Apurv.Mishra@amd.com>
Honglei Huang
Tony Gutierrez
Yiannis Papadopoulos
Ken O'Brien
Sunday Clement
David Yat Sin
Chris Freehill
Alysa Liu
Apurv Mishra