- Avoid GPU wait on the marker submission and update the command
batch after HSA signal callback upon HSA barrier completion.
Change-Id: I5c1c97212aefc2ae4b99aa9e2a81627ee9a38c1c
In pitch workaround, we need to copy the image to copy buffer
when application wants to read image to buffer. After this
patch, we unconditionally copy the image data to the copy buffer.
Change-Id: I71b0d19459542dfbb3ca51a2c8a3a81367fa2fb5
The existing workgroup calculation logic for GWS initialization is
incorrect. It tries to add together workgroups across dimensions,
leading to major under-count in 2D and 3D kernels. An (x,y,z) kernel
uses x * y * z blocks, not x + y + z.
In addition, the previous logic was incorrect for the case of launching
a single-threaded kernel. It calculated 0 workgroups, leading to
initializing GWS to -1.
Change-Id: I1bb20a0d5b6e0cc10ac55901c28d8f93aac61c09
- The logic will trace compute, sdma read/write operations and
apply signals when necessary
- ROC_CPU_WAIT_FOR_SIGNAL, ROC_SYSTEM_SCOPE_SIGNAL
and ROC_SKIP_COPY_SYNC were added to control the tracking
Change-Id: I9e8e6174c63bf7784f7ab00964e2918c8667d364
- ROCR fails the call for some reason, then the signal will
become invalid and can hang on a wait. The logic will reset the
active signal in such cases
Change-Id: Ia131420200f1bbd7c9a162b8f1b06db8cecf41c6
- There is a performance regression with a HW wait for HSA signal
on ROCr async operation. For now move the logic back to CPU wait.
- Fix profiling issue with multiple HSA signal per single timestamp
object. Some copies require multiple ROCR calls and if profiling is
required, then the execution time is derived from all used signals.
Change-Id: Id003e4abb8c2de378eedc152a7e389500fc6f4ce
Remove targetId_, gfxipMajor_, gfxipMinor_ and gfxipStepping_ from
device::Info as they are now available in device::Isa.
Change-Id: I381b1d4798ebf50655740e004a01ac7f86dbf668
- Correct GSL path to report targets using the TargetID syntax.
- Correct GSL path to check compatibility of code objects when
loading.
- Add concept of an device isa and create a registery used by ROCm,
PAL and GSL.
- Support XNACK and SRAMECC target features consistently for PAL and ROCm.
- Correct logic for NullDevices and asserts to avoid memory coruption.
- Allow all NullDevices to be created for HIP.
- Numerous other code improvements.
Change-Id: I40abf3d2b22249c1492d1af5919665f8184f4e0e
Use strncpy instead of strcpy to ensure the arrays will not be
overflowered. Only copy one less than size of char array to leave a
NUL character at the end even if the copy is truncated provided the
original object is zeroed memory.
Change-Id: I00f7679630cf28dcb9a51cb0aba2810a4f4c72b9
For roc devices create hsa_agent_t handles using a pointer to the
device::Device. This ensures each device has a different hsa_agent_t
handle. This may be necessary to ensure the loader symbol lookup will
search only symbols for the correct device.
Change-Id: Iee6dd40d68bf22a02ce8c75cbe5ac8f5a0d9e418
roc:Settings and pal::Settings are derivations. Allocate them as their
derived class then assign that to the base class member to avoid the
need for a static_cast.
Use device::settings to access the Settings consistently.
Change-Id: I0f85157962fbf6fed176da0caf83b723bcbe1452
Make roc::Device::getBackendDevice non-virtual and only defined by
roc::Device and not roc::NullDevice. It is only meaningful for an
online device.
Change-Id: Ic333a3a42f650bea524e80dab587a34f1353e593
Delete roc::Program::hsaDevice and access directly from device
associated with program. This allows to be clear if the device is a
NullDevice which has no meaningful HSA agent backend device.
Change-Id: I81f96aff47bf9b8166d0ff6a5efc7c01f0fb6de3
Make roc::Device constructor and create() method private as creating
devices is performed by the factory static methods.
Change-Id: Ifa2edb8ec645b4ce6070c4aef355b9ef88294cf1
- Add assertions to enforce that objects are of the correct kind and
have been allocated.
- Make destructors check if objects have been allocated before
deleting.
- Operations that require a non-NullDevice return failure if given a
NullDevice.
- Use static_cast rather than reinterpret_cast when cohersing from a
base class to a derived class.
Change-Id: I02ee0ea9d7982fd7ca29d49c9b02cfae111b7127
German Andryeyev
Vladislav Sytchenko
Chauncey Hui
Rahul Garg
cjatin
Julia Jiang
Jason Tang
Alex Xie
Joseph Greathouse
Saleel Kudchadker
kjayapra-amd
Jason Tang
Tony Tye