Some libs use __HIP_NO_HALF_OPERATORS__ and __HIP_NO_HALF_CONVERSIONS__
which results in operators being hidden and can cause errors.
Change-Id: I83c194d7d727cba30b46d7c296f7d396549f5fca
In active wait mode use signals without interrupts by default and switch
to the interrupts only if a callback is required.
Change-Id: Ibcde8f7d44c70f8fb8fa5e0a7fdd8b08a2982a8e
On Windows, hipHostRegister may add a single object in the MemObjMap
that maps to memory that is allocated on different devices.
This change ensures that the offset that is returned from
getMemoryObject() is computed relative to the memory that is allocated
on the current device.
Change-Id: I5fd3af200bf6f4926fdeaea12dcb9d0154d3a843
- Header files inside rocclr/utils when included from hipamd or opencl should be included as #include "rocclr/utils/xxx.h" instead of "utils/xxx.h"
Change-Id: Ic0760c33b9d091f5620dec67e5482c9698d22093
Some functions were __device__ only, but should be __host__ and
__device__, changed them to __HOST_DEVICE__.
Some functions were __HOST_DEVICE__ but were using ockl functions,
changed them to __device__ only.
Change-Id: Ife9e7abe60415bda68f5f9a101e6e7c39ad51064
The hipGraph will use VMM by default when allocating memory.
However, the handle of Phy mem has been added to Memobj by default.
Since the Memobj will track the whole address range from handle to
handle + size, this needs the system to reserve the whole address
range. If the system range have not reserved by the system, then it
will have the potential issue that clr finds the Memobj incorrectly.
This patch removes the handle from the Memobj to fix this potential
issue.
Change-Id: I2da38e6b2d11d0d48e1afe66c46899500c290624
- Refactor blit code and clean ASAN instrumentation
- Use unified function for rocr copy
- Enable shader copy path for unpinned writeBuffer/readBuffer paths
- Set GPU_FORCE_BLIT_COPY_SIZE=16 which means we will use BLIT copy for
pinned copies or unpinned H2D/D2H copies < 16KB
Change-Id: I42045cca79234b340dbf53dafb93044199736ae4
The early return if the thread is not alive causes memory leaks.
Neither doorbell_ or urilocator are released if the thread is not alive.
This change alters the logic so regardless of the thread status the
HostcallListener releases its memory.
Change-Id: Ie912360ec0e2ee257de9937b1a8d7375e6aebd83
This change replaces some asserts, that were only available in debug
mode, with standard error handling.
Signed-off-by: Sebastian Luzynski <Sebastian.Luzynski@amd.com>
Change-Id: I112f9e56f921abd72daf0d11e4ecdcb7b1a9f9e6
=> If null stream is not created during sync skip nullstrm creation
=> Do cpu wait on blocking & null stream if it exists
Change-Id: I90d6ced6a2dd1782ba58f3fed4e3608fc0efa55a
hipMallocAsync/hipFreeAsync APIs should return error stating
operation is not supported, if a stream is actively capturing
and is different from the passed stream
Change-Id: I2a1b8260c5eb22d99a936ac529d6788a83f81a17
1) for case where kernelParam_.func = nullptr, the validation
fails in setParam call and memory is not alloced for kernelParams
2) destructor path segfaults trying to free the kernelParams memory.
3) copy of params is done after function validation is successful.
Change-Id: I6338e0c89f259632e4115f0508e2f240bc207fd9
hsa_amd_profiling_async_copy_enable is taking 45us for the first call. Disable sdma profiling for enqueuing captured kernel packets and for accumulate command.
Change-Id: I80b51a58c46bccc9c1025e9331515f57c97b5a2a
Anusha GodavarthySurya
Jaydeep Patel
Saleel Kudchadker
Maneesh Gupta
German Andryeyev
Jatin Chaudhary
Todd tiantuo Li
Gerardo Hernandez
Ranjith Ramakrishnan
Ajay
Branislav Brzak
Ioannis Assiouras
Marko Arandjelovic
Shane Xiao
Jimbo Xie
Sourabh Betigeri
Aidan Belton-Schure
Sameer Sahasrabuddhe
taosang2
Vladana Stojiljkovic
Sebastian Luzynski
Marko Arandjelovic
Satyanvesh Dittakavi
amd-jmacaran
Rahul Manocha
kjayapra-amd