Remove a workaround to CS_PARTIAL_FLUSH added in CL#1495187,
since PAL is no longer uses CS_PARTIAL_FLUSH.
Change-Id: I03edc7595459e19aad33b2b0901f0ebe4754d310
[hipclang-vdi-rocm][perf]~45% to 50% of Performance drop on
rocBLAS_int8 test
- Enable AMD_OPT_FLUSH optimization by default to match HCC
- Disable CPU writes to GPU memory on boards with large bar,
because it requires HDP flush tracking.
- Enable L2 cache on kernel arguments, because L2 will be
invalidated on memory reuse .
Change-Id: I124cf250bdd4d19c523ce542c163813828f8fbdc
The existing one can have issues on certain systems, therefore this limits use of direct memcpy via largeBAR to sizes where it is unequivocally better.
Also addresses SWDEV-220030 and SWDEV-222237.
And also don't optimize the case where start==stop event to compute
elapsed time since the command can be a NDRange one.
HIP directed test will need to be fixed for that.
Change-Id: I64fadd6ab8ab1a490e7a2b7165a591df5a5cf3a2
This is a quick workaround to match HCC behavior for performance since inlining usually
results in more optimization opportunities therefore better performance.
We will fine tuning inline threashold later.
Temporarily comment out Hcc-specific template functions
hipExtLaunchKernelGGL and hipOccupancyMaxPotentialBlockSize for CLang
compiler so that all test cases under hip/samples can be built
successfully for Clang + Hip/Hcc runtime.
Change-Id: Iafc761257be4a7b34eafa6759a01f369570cd6ce
For hipLaunchKernelGGL(), hmod->kernargs is empty, thus we need
insert hmod->kernargs[name_str] which is empty.
Change-Id: I95f818d0525da84452e66c5778f0648a643843c7
Fix the following issues:
1.Ignore hidden arguments of kernel functions.
2.Look up both origial function name and function name with .kd postfix
when argments are retrived from module.
3.Addition, fix compiling issue of LaunchKernel test app.
Change-Id: I9400943f2f02433cb4409b19c0cac3626c2bc454
* Use deque instead of vector for code readers so that the iterators and references will be stable
* Fix compile error
* Assign the iterator
* Add multithreaded test
* Make threads a multiple of hardware concurrency
* Output on failure
* Add setDevice to try and initialize the context on cuda
* Create context for cuda
* Set context on each thread
* Reduce threads on cuda
* Skip test on cuda
* Try to initialize the primary context on cuda
* Push ctx to the stack as current
* Revert "Push ctx to the stack as current"
This reverts commit e8a7cc6957.
* Revert "Try to initialize the primary context on cuda"
This reverts commit 86cdb30195.
* updated test for nvidia path
* Add c++11 option for nvcc
Co-authored-by: satyanveshd <53337087+satyanveshd@users.noreply.github.com>
Some PyTorch unit tests have regression. Disabling cov3 to allow more
time to debug and unblock PyTorch
Change-Id: Iba7f425ef3499c20c42ec45d9152b5d27ce97d03
Karthik Jayaprakash
Jason Tang
Christophe Paquot
Payam
Jatin Chaudhary
agodavar
Sarbojit2019
eshcherb
Tao Sang
Evgeny Shcherbakov
German Andryeyev
Vladislav Sytchenko
Alex Voicu
Yaxun (Sam) Liu
Rahul Garg
Maneesh Gupta
Saleel Kudchadker
Nick Curtis
Siu Chi Chan
Evgeny Mankov