* Continuous integration
* linux-ci on
* fix for parallel-overhead
* Updated CMAKE_INSTALL_PREFIX
* Updated installed boost libraries
* Dyninst updates fixing TBB install
* timemory + dyninst submodule updates
- fixes some timemory package option handling
- fixes dyninst libiberty handling
* Update dyninst submodule with libiberty fix
* dyninst submodule update with TBB internal build fixes
* Updated linux-ci + tests + timemeory + dyninst
- updated timemory submodule
- update dyninst submodule
- delay OnLoad implementation
- DYNINST_RT_API handling improvement
- CI for ubuntu-bionic in addition to focal
- CTest in CI
* Update TBB handling in CI
* Update dyninst with symLite fix
* Update dyninst submodule with ElfUtils-External fix
* Dyninst::ElfUtils fix
* Modified dyninst submodule TPL install
* Update dyninst submodule with improved interface libraries
* Fix to Dyninst::ElfUtils in dyninst submodule
* Updated CI build matrix + test install
* Updated CI
* CI stage updates
* Tweak
* Dyninst updates + RPATH for hosttrace exe
- hosttrace will rpath to dyninst
- Dyninst will statically link to boost by default
- Fixes to double init w/ MPI + runtime instrumentation
- minor cleanup to hosttrace
- hosttrace help exits with zero
- CI updates
* Dyninst + CI updates
* Removed ELFUTILS_BUILD_STATIC option in Dyninst
* Dyninst + visibility + roctracer + tests
- Dyninst submodule updates with dynamic pcontrol lib
- hosttrace visibility is hidden
- improved handling of DYNINST_API_RT in hosttrace
- roctracer::tear_down in finalize fixes issues with timemory
- throw error if cannot create perfetto output file
- roctracer_default_pool() safety guards
- resume calling roctracer_close_pool()
- fixed working dir of tests
- fixed env of tests (cmake and CI)
- simplified CI
* Removed stray CI if condition
* Disable hidden visibility for now
* ifdef for roctracer::tear_down + reenable hidden
* Better CI environment handling + dyninst packaging
* Fix for dyninst-package CI
* Fixes for cmake 3.15 issues with aliasing imported lib
* Fix to ubuntu-focal-dyninst-package
* Dyninst updates for packaging
- fixes issues with hard-coded paths to libraries after relocation via package installer
* Restrict CI to main and develop branches
This was never validated on GFX10, so remove it for now until it's
either validated and confirmed, or move the check to the test itself
Change-Id: Ie4d8b31885fbe6e5ed84b7b174c0bfed60879741
to faciliate debugging, print errno when queue destroy fail
current log give very little information when fail:
[ RUN ] KFDQMTest.AllSdmaQueues
[ ] Regular SDMA engines number: 1 SDMA queues per engine: 2
[ OK ] KFDQMTest.AllSdmaQueues (11 ms)
[ RUN ] KFDQMTest.AllXgmiSdmaQueues
[ ] XGMI SDMA engines number: 0 SDMA queues per engine: 2
[ OK ] KFDQMTest.AllXgmiSdmaQueues (6 ms)
[ RUN ] KFDQMTest.AllQueues
/home/foreman/build/hsakmt-roct-amdgpu-1.0.9.40500/sources/libhsakmt/tests/kfdtest/src/KFDQMTest.cpp:381: Failure
Value of: (cpQueues[i].Destroy())
Actual: 1
Expected: HSAKMT_STATUS_SUCCESS
Which is: 0
Signed-off-by: Yifan Zhang <yifan1.zhang@amd.com>
Change-Id: I5b1b5b616a5fd7ff198360c893a7aeed685022bd
Instead of detecting files (header/library), use cmake's find_package to
locate the required dependencies (hsa-runtime64 and hsakmt).
Adding hsa-runtime64::hsa-runtime64 and hsakmt::hsakmt to the
target_link_libraries also takes care of adding the interfaces include
directories to the search path.
Change-Id: I64eb77c97dac7982ac96d3158ad57df776cc0b53
For minimal latency we should place command queues and blit code
in the nearest numa node to each GPU. Add an allocator matching
the current runtime default allocator interface to each GpuAgent
that allocates on the closest numa node as represented by kfd
topology. Use this allocator for queue ring buffers and blit
objects.
Change-Id: I181127f9c27bafe68976312963146616e3f58369
After calling ioctl to create userptr obj, take aperture lock, check if
there is same userptr obj created after finding object, to catch the
race that multiple threads register same userptr to multiple GPUs.
If same userptr obj exist, then increase userptr registeration_count,
and free the newly create obj.
Change-Id: I63ae3a4f54da8aedd11c124d8d53ebe727b8203a
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Map readonly range to GPU, test GPU can read range but write to
range trigger GPU vm fault.
Change-Id: Id2ce106055a353f21f4c34fa7f562d546523bc49
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
To test if SVM support address mmap on file backed memory.
Change-Id: I227cc0b44d167c4fa2e63257e13a481aefa4f750
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
- hosttrace library automatically collects and merges timestamps for HIP API calls and kernels with the host-side instrumentation
- mostly eliminates the need for using external rocprof
- added thread_instruction_count in perfetto output
- increased hosttrace min_loop_address_range to 512
- disabled instrumenting functions with dynamic callsites by default
- miscellaneous cmake updates
* roctracer support
- fully integrated perfetto + roctracer outputs
- thread_instruction_count in perfetto
- increased min_loop_address_range to 512
- disabled instrumenting functions with dynamic callsites by default
- updated timemory submodule
* hosttrace_launch_compiler
- support for using an alternative compiler as needed via launch compiler
- elfio added as submodule (not currently used)
- miscellaneous cmake updates
* README update + host/device categories + misc
- timemory fix for TIMEMORY_ROCTRACER_ENABLED
- transpose fix
* papi_tuple_t -> papi_tot_ins
- minor fix to Findroctracer.cmake
L2 flush is triggered by explicit cache flush PM4 packet in aqlprofile
packets to GPU. This cache flush is used to sync up CPU and GPU to make
sure perfomance counters copied to profile output buffer is visible to
CPU. To get rid of this cache flush the followings are done:
1. This explicit cache flush packet is removed from aqlprofile code
(another commit to aqlprofile code).
2. This commit which changed profile output buffer to use kernarg
memory since it is uncached for GPU.
After these changes profile counter values when copied by GPU to output
buffer they are guaranteed to be visible to CPU.
Change-Id: Ie953949c85fbee2f4369f1de966bcfb33daec084
- tweaked some CMake option names
- moved merge-trace.jl to hosttrace-merge.jl
- removed Windows line encodings from hosttrace-merge.jl
- improved handling of !perfetto and !timemory
* various tweaks
* build updates + cleanup + overlap guard + min addr range
* Library source reorg + miscellaneous tweaks
* Removed unnecessary fwd decls
* Print address range in --print-X pair mode
- hosttrace modifications
- disable instrumenting functions with overlapping sections or multiple entry points by default (control via --allow-overlapping option)
- disable instrumenting functions whose address range < 512 bytes unless a loop is present by default (control via --min-address-range option)
- disable instrumenting functions w/ loops whose address range < 64 bytes (control via --min-loop-address-range)
- Support for wrapping MPI function calls even in binary rewrite mode
- e.g. use gotcha to wrap MPI functions with hosttrace_push_trace and hosttrace_pop_trace
- New timemory only mode --> HOSTTRACE_USE_TIMEMORY=ON
- New timemory + perfetto mode --> HOSTTRACE_USE_PERFETTO=ON + HOSTTRACE_USE_TIMEMORY=ON
- Full support for all timemory components
- parallel-overhead example for measuring the overhead in a MT-parallelized application with very small instrumentation functions
- improvements to output directories for hosttrace exe
- improvements to output directories for hosttrace library
- new hosttrace options
- --print-instrumented <type> prints out the instrumented entities and exits
- --print-available <type> prints out the available instrumentation entities and exits
- --print-overlapping <type> prints out the overlapping entities and exits
- NOTE: <type> above refers to the information printed out, e.g. module name vs. function name vs. module and function name, etc.
g_props is intialized in topology_take_snapshot which needs to call
hsaKmtAcquireSystemProperties. In hsaKmtOpenKFD, it doesn't call hsaKmtAcquireSystemProperties.
So it needs to change parameter of topology_is_svm_needed from node_id
to EngineId to avoid Segmentation fault on gfx902.
Signed-off-by: changzhu <Changfeng.Zhu@amd.com>
Change-Id: Iba8d20a9142510c70927454d26bdcaf579ad5574
Also make failure to handle queue errors fatal.
Motivation is to improve detection of queue error conditions
that currently appear as application hangs.
Change-Id: I655643616dc0bd303d7df3ce8aca2c099bec3d46
Sets package found and component lists. ROCr does not have components
so this is mostly cosmetic. It's part of maintaining a compliant
cmake project config file though.
Change-Id: Ida2ef746375143babd3a6f938727a47135606f01
Per clang 13 option -Wno-error=unused-but-set-variable is not
recoginized nor is the diagnostic emitted. Set this option
conditional to the clang compiler version.
Change-Id: I3c0958dffa985d53b641f9eff4e702988dffd033
Passing 0 into num_cu_mask_count used to be an implicit error.
This has been repurposed as a short hand for enabling all CUs.
Enabling all CUs when HSA_CU_MASK is set will cause the CU mask to
reset to whatever was set by HSA_CU_MASK which may then be queried.
Change-Id: I1d6bb2034595a78ee48fa72aa05563e8ea6c0fff
Delay parsing until after GPU discovery. Use the surfaced
GPU count and maximum phyiscal CU count to limit parsed bit masks.
This prevents pathological input such as
HSA_CU_MASK=0-8000000:0-8000000 from attempting to consume 7TiB.
Change-Id: I3773d2db3740c2023b0f6275d1818b69119b0495
Use the database provided by libdrm-amdgpu,as this is maintained by AMD
Signed-off-by: Mike Li <Tianxinmike.Li@amd.com>
Change-Id: I319a46d833cb23173a77fdff1deae69ea58588b8
Consolidate all thunk components into a single development package
Also declare build dependencies via find_package/library and report the
location of said dependencies during build time
Change-Id: I8524ad1a2d9a664accd93471abacc4bac82b9a7c
Signed-off-by: Kent Russell <kent.russell@amd.com>
Update KFDCWSRTest.BasicTest to run CWSR test with legacy HWS,
MES, and software scheduler.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Change-Id: I09af904507109c2bbbace6ff616a57fa2b1b8aa9
Jonathan R. Madsen
Saurabh Verma
Kent Russell
Yifan Zhang
Oak Zeng
Sean Keely
Laurent Morichetti
Philip Yang
Chun Yang
changzhu
rachida
Chengming Gui
Jinzhou Su
Jianbing Chen
Mike Li
Elena Sakhnovitch
Jianbing Chen
Tony Tye
Mukul Joshi