* SWDEV-550626 - Refactor atomics header and tests
1. Introduce __HIP_ATOMIC_BACKWARD_COMPAT.
By default we define __HIP_ATOMIC_BACKWARD_COMPAT=1 to
let hip atomic functions maintain old assumptions. if
users want to adopt the new behavior, that is , by default
assume no-fine-grained no-remote-memory, then they can
define __HIP_ATOMIC_BACKWARD_COMPAT=0 and get the new
behaviour.
2. Use __HIP_ATOMIC_BACKWARD_COMPAT_MEMORY to replace
original __HIP_FINE_GRAINED_MEMORY in atomic header.
And apply __HIP_FINE_GRAINED_MEMORY onto all
atomicXXX_system() functions to prevent failure on memory
allocated by hipHostMalloc().
3. Replace HIP_TEST_FINE_GRAINED_MEMORY with
HIP_TEST_ATOMIC_BACKWARD_COMPAT_MEMORY in hip-tests.
4. Fix negative test errors.
Fix managed memory test error on memory order.
some other minor changes.
As a result all originally disabled tests are enabled.
5. Add more atomics tests in some cases.
6. Reduce test time in each case.
Reduce iteration number to 1 for tests that cost too much time.
8. Put common codes into hip_test_common.hh
Return error when ext_fine_grain_pool is unavailable for
hipHostMallocUncached, hipHostAllocUncached and
hipExtHostRegisterUncached.
Disable related tests on Navi4x where
ext_fine_grain_pool is unavailable
* SWDEV-523177 - Fix the Unit_hipEventCreateWithFlags_DisableSystemFence_CohHstMem test
* Fix the hip nvidia psdb failure
[ROCm/hip-tests commit: 41af8e62ca]
1.Remove clock functions from some tests that don't need them.
2.In some memory pool tests and coherency tests, timer-based kernel
delay isn't reliable, use pinned host based notification instead.
3.Add CHECK_PCIE_ATOMICS_SUPPORT before some tests.
4.catch/unit/memory/hipMemoryAllocateCoherent.cc is removed
as it is useless and originally excluded in building.
5.Some tests can still pass even if clock rate =0, thus they
will be kept as is.
6.Some logic and format improvement in some tests.
Change-Id: I6b3c6bf54c61cffd45cd6f17c75998f751b75725
[ROCm/hip-tests commit: ec8ff45a1d]
- For H2D transfer cases, we cannot enfore ErrorNotReady behavior if the
source is unpinned. In that case the driver can sync depending on if
it pins or stages the source buffer.
Change-Id: I436a4b667bff172cfbadc58631a46a215c40fcb1
[ROCm/hip-tests commit: d8f5c2560f]
- Since any-host-to-any-host as always synchronous, there's no need for
separate functions
- Removed 'return' statement as it skips the rest of the test.
Change-Id: If70e0b21ad8151311d5ac7c5de789f6265cf44b6
[ROCm/hip-tests commit: 3809d1bb75]
Fix memory leaks by adding missing destroy calls for
events, streams, and graphs at the end of tests.
Ensure that every test case executes destroy calls,
regardless of whether it passes or fails.
Change-Id: I814e35c528d90ed2abb34d77377f1a7fd3f1f11c
[ROCm/hip-tests commit: 9cffda4ebb]
* _sync tests that use warp match functions (__match_any_sync and __match_all_sync) were disabled for nvidia because those functions are supported only on devices with compute capability 7.x or higher. In this solution, tests are always build (with specifying appropriate compile flag) and in in the runtime it is decided weather they should be skipped or not.
Change-Id: Ifa9e55f3b47aa3a00027cb986cd6ae46aed45ebd
[ROCm/hip-tests commit: 9893726343]
Add kernel excution perf test per blockSize and block number.
Implement a solution to roughly evaluate gpu variable
frequency based on clock64() and wall_clock64().
Change-Id: Ic87761a862d4a894fdcaab3431d63fe2592bb682
[ROCm/hip-tests commit: 0a22d14775]
Modified
TestContext::finalizeResults()
Added
HIP_CHECK_IGNORED_RETURN()
HIP_CHECK_ERRORS()
to supported ignored error such as hipErrorNotSupported.
Enable all mipmap tests in catch2 memory folder for linux.
The issues in the following tickets should be fixed.
SWDEV-465162, SWDEV-465163, SWDEV-465165, SWDEV-465166,
SWDEV-465168, SWDEV-465171, SWDEV-465172
Change-Id: I1146d6184582a7111a5af25b52f9e12c6a160e49
[ROCm/hip-tests commit: de9581efb6]
The following builtins are tested:
- __all_sync, __any_sync, __ballot_sync and __activemask
- __match_any_sync and __match_all_sync
- __shfl_sync, __shfl_up_sync, __shfl_down_sync, and __shfl_xor_sync
The tests for shfl (all variants) were manually validated on a CUDA setup.
NOTE:
- Unit_hipVoteSync_All temporarily disabled on Windows (SWDEV-452308).
- All new tests temporarily disabled on CUDA (SWDEV-453145).
Change-Id: I84b205a88aa24219d199c760793e2f19f0cf8f13
[ROCm/hip-tests commit: 53f0a9bd01]
This change adds a new cmake target to generate test plan documentation using
Doxygen. The target is called build_docs and it's not added to the default
build commands, hence it needs to be called separately.
Signed-off-by: Sebastian Luzynski <Sebastian.Luzynski@amd.com>
Change-Id: I441524568e0513019ca21467950e48a551234802
[ROCm/hip-tests commit: 53406684ee]
Multiple subtests will fail if the default mem pool will keep information
of the previous subtests.
Change-Id: Ic9ab8ed93771e71735594a85235b5bb734d6680c
[ROCm/hip-tests commit: 4e3fc2d937]
MachineTom
vstojilj
systems-assistant[bot]
Julia Jiang
Danylo Lytovchenko
Dittakavi, Satyanvesh
Luo, Phoebe
Manocha, Rahul
GunaShekar, Ajay
Belton-Schure, Aidan
Hernandez, Gerardo
Saleel Kudchadker
Marko Arandjelovic
Rambabu Swargam
Vladana Stojiljkovic
Li, Todd tiantuo
SrinivasaRao
Ioannis Assiouras
sdashmiz
Sameer Sahasrabuddhe
Jaydeep Patel
Sebastian Luzynski
German Andryeyev
Rakesh Roy
Mirza Halilčević
Mirza Halilcevic
Nives Vukovic
Milos Mozetic