When HSA_USE_SVM is 0, thunk uses non-SVM path, but upper layers still
use SVM path. That is not as expected.
Suggested-by: Lang Yu <Lang.Yu@amd.com>
Change-Id: I1ae0b4faa2f8af5ec69a81cfeb7661bd47d739d4
As the KFD IPC IOCTLs will not be upstreamed, change runtime
implementation to use DMA bufs.
DMA buf fds will be passed over abstract unix domain sockets.
The exporter spins a thread that creates a socket server.
The importer connects to the server to fetch the fd.
libDRM will be required to do a manual import and GPU map for
memory that is not already imported and mapped.
For now, use the legacy IPC implementation by default as a
follow on patch will disable the HSA_ENABLE_IPC_MODE_LEGACY
environment variable.
Change-Id: Ifd8469e9adfc81f8a1ea78d6010fb10b515ba1b4
Implement HybridMutex to improve latencies compared to KernelMutex when
there is contention between several threads calling hsa_signal_create
and hsa_amd_signal_async_handler.
Change-Id: If53377033e749b0050727964c9303f09b02527cc
Fixes issue where t1_ counters may not be updated when doing dispatch
profiling, causing a divide by 0.
Change-Id: I91060ac3f9fd2183d277e6e7cd810398a453a87f
KFD had some fixes for handling of virtual memory APIs. These fixes are
included in interface version 1.15.
Change-Id: Ie701eccf6e032f9ec0a1f4e8a43718964eebddc6
Update the `hsa.h` header to use the gcc / clang `__BYTE_ORDER__`
macros where available to more accurately autodetect endianness for
the target.
Change-Id: I7312f3badcba9287a30eb14882b91e2a247acc5f
This reverts commit c5db063b2f. This
change is required for the runtime to generate reliable core dump files,
but this feature has been disabled for now by
5e3be9c28a. Until it is needed, revert
the ABI change in the trap handler to maintain compatibility with older
debugger.
Change-Id: I77a1562dc7962befe2bf88442df858e2d2b1c5ab
- librocprofiler-sdk.so no longer directly links to HIP runtime library (libamdhip64.so)
- librocprofiler-sdk.so no longer directly links to HSA runtime library (libhsa-runtime64.so)
- librocprofiler-sdk.so does directly link to AQL profile library (libhsa-amd-aqlprofile64.so) so HSA runtime library is still in DT_NEEDED
- This needs to be fixed eventually so rocprofiler-sdk library can be independent of HSA and HIP runtimes
* Tests for agent and aql packet generation
Test for agent and fixing test problems with aql packet that caused test
to not run.
* cmake formatting (cmake-format) (#366)
Co-authored-by: bwelton <bwelton@users.noreply.github.com>
* source formatting (clang-format v11) (#367)
Co-authored-by: bwelton <bwelton@users.noreply.github.com>
* Minor tweak
* source formatting (clang-format v11) (#368)
Co-authored-by: bwelton <bwelton@users.noreply.github.com>
* Add gfx900 to basic_counters
* Update samples/counter_collection/client.cpp
- fix data race by flushing buffer during tool_fini
* Fix data race for output stream destruction
---------
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: bwelton <bwelton@users.noreply.github.com>
Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com>
* Fix init order for getMetricIdMap
Ensure that getMetricIdMap() stays valid until the destruction
of the counting service. Adds a test to ensure that this behavior
is maintained (this test fails without this change).
* source formatting (clang-format v11) (#343)
Co-authored-by: bwelton <bwelton@users.noreply.github.com>
* cmake formatting (cmake-format) (#342)
Co-authored-by: bwelton <bwelton@users.noreply.github.com>
* Remove threading
* Update usage of rocprofiler::counters::get{Metric,MetricId}Map()
- uses common::static_object
* Update counter init_order testing
---------
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: bwelton <bwelton@users.noreply.github.com>
Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com>
Apply the following changes to project documentation for ReadtheDocs:
add version number to documentation left navigation bar and page title
add an "About" section with a license page
enable htmlzip, pdf, epub formats when publishing on Read the Docs
set pdf title, author, copyright, and version
rename .sphinx/.doxygen to sphinx/doxygen
remove docBin from URL
update rocm-docs-core dependency
update dependabot config
Change-Id: Ife8c89a2e9323f436b3e54ef2a9e013c19b3b228
* Fix omnitrace-avail component list
- remove omnitrace components from `omnitrace-avail -C` since these are no-ops in OMNITRACE_TIMEMORY_COMPONENTS
* Fix omnitrace-avail-filter-wall-clock-available test
* Update build-docker.sh
- support rocm 6.0
* Update cpack workflow
- support rocm 6.0
* Update CI testing workflow paths-ignore
- changes to {docs,cpack,containers,formatting}.yml and docker do not require testing
* Update docker for OpenSUSE
- always use --non-interactive with zypper
- tweak to PERL_REPO when OS version >= 15.4
* Rename OMNITRACE_USE_PERFETTO to OMNITRACE_TRACE
* Rename OMNITRACE_USE_TIMEMORY to OMNITRACE_PROFILE
* Revert change to Perfetto.cmake
* Fix formatting
clang-format-11 was complaining about formatting
* Update lib/core/config.cpp
- Add OMNITRACE_ROCTRACER_HIP_API_BACKTRACE option
* Update lib/omnitrace/library/roctracer.cpp
- support perfetto debug annotation of backtrace in HIP API call
* Fix backtrace resolution and ordering in UI (#1)
* Fix backtrace resolution for non-omnitrace libraries
* Nicer Perfetto UI on long backtraces
* Make Perfetto annotation consistent
* clang-format
---------
Co-authored-by: Tal Ben-Nun <tbennun@users.noreply.github.com>
* Fix roctracer_flush_activity
- invoke roctracer_flush_activity() before disabling domains
* create comp::roctracer::flush()
- real issue was the global state when roctracer_flush_activity() was called
* formatting
* Update lib/omnitrace/library/components/roctracer.hpp
- provide definition of comp::roctracer::flush when OMNITRACE_USE_ROCTRACER is not defined
* omnitrace.cfg -> perfetto.cfg
- rename provided perfetto config file (omnitrace.cfg) to perfetto.cfg to avoid confusion
* Update lib/core
- gpu.hpp: defines for OMNITRACE_USE_{HIP,ROCTRACER,ROCPROFILER,ROCM_SMI}
- gpu.cpp
- include core/hip_runtime.hpp
- fix serialization of hipDeviceProp_t
- add hip_runtime.hpp
- ensure proper inclusion of hip_runtime.h
- add rccl.hpp
- ensure proper inclusion of rccl.h
* Update lib/omnitrace/library
- rcclp.cpp
- update includes for rccl
- roctracer.hpp
- update includes for hip_runtime
- components/comm_data.hpp
- update includes for rccl
- components/rcclp.hpp
- update includes for rccl
* Update bin/omnitrace-avail/avail.cpp
- update includes for hip_runtime
* Update examples/rccl/CMakeLists.txt
- fix find_package for rccl when CI enabled
* Update CMakeLists.txt
- set cmake policy CMP0135 to NEW for cmake >= 3.24
- Enable DOWNLOAD_EXTRACT_TIMESTAMP with ExternalProject_Add + URL download method
* Update timemory submodule
* Update pybind11 submodule
* Update pybind11 submodule
* Update lib/core/rccl.hpp
- include rccl.h only if OMNITRACE_USE_RCCL > 0
* Update lib/core/{gpu,hip_runtime}.hpp
* Update lib/core/gpu.cpp
- reintroduce some ppdefs
* Update lib/core/gpu.cpp
- fix ifdef on OMNITRACE_HIP_VERSION
* Update lib/core/gpu.cpp
- fix static assert for OMNITRACE_HIP_VERSION_MINOR when HIP version 4.x or older (unreliable minor versions)
* Update lib/core/gpu.cpp
- fix ifdef on OMNITRACE_HIP_VERSION
* Update lib/core/config.cpp
- disable OMNITRACE_PERFETTO_COMBINE_TRACES by default
* Update lib/core/perfetto.cpp
- if unable to open perfetto temp file, return the ReadTraceBlocking()
* Update lib/core/config.*
- flush tmpfile before closing
Also add stddef.h workaround for old GCC.
RHEL-8 still uses GCC 8.5 and templates are not well supported.
Change-Id: Ia4dae23892ec63682ea848c46ba81de85cf6d209
Signed-off-by: Galantsev, Dmitrii <dmitrii.galantsev@amd.com>
NOTE: RVS Build is disabled by default due to CI build issues.
Change-Id: I1593f0fe22075a9f86f54afa3ac151e109f1f7bd
Signed-off-by: Galantsev, Dmitrii <dmitrii.galantsev@amd.com>
-m64 only applies to x86-64, it's more reasonable to enable it only on x86_64 host.
It fixes build on some other platforms as well.
Provided by user r-value on github
https: //github.com/RadeonOpenCompute/rocminfo/pull/63
Change-Id: I9c1c40d3fa39b0a61d28041fe4998b5e1ad0cdcd
Adds support and implement APIs for 'gpu_metrics_v1_5'
Code changes related to the following:
* gpu metrics 1.5 support
* Unit tests
* Examples
Build changes related to the following: None
Change-Id: Ie8917dd63c1dd1a94467b100fa44b634cebe62b6
Signed-off-by: Oliveira, Daniel <daniel.oliveira@amd.com>
Giovanni LB
Jonathan R. Madsen
Jonathan R. Madsen
Benjamin Welton
David Yat Sin
Jonathan Kim
Joseph Huber
Lancelot SIX
coleramos425
Galantsev, Dmitrii
Sam Wu
guanyu12
SrirakshaNag
Ben Richard
Tal Ben-Nun
Gopesh Bhardwaj
Ranjith Ramakrishnan
Oliveira, Daniel