* Fix cooperative launch APIs to set hipGetLastError
Previously, the cooperative launch APIs did not properly log their
errors in the global hipGetLastError variable before returning back
to the user. As such, the APIs would leave hipSuccess in the
last error, which would break some use cases.
This fixes that problem by making a trampoline function that does
the HIP_INIT_API and ihipLogStatus.
* Add missing flag to the log of multi-GPU launch
[ROCm/clr commit: 341ef7fdca]
This PR takes ensures that the maxThreadsPerBlock returned by hipFuncGetAttributes is both a multiple of the warp size and that the register usage of the maximum block does not exceed the number of available registers.
Fixes#1662
[ROCm/clr commit: ea28d64297]
Query ROCr to see if we have the proper lower-level support for
cooperative groups -- GWS support through the firmware, driver,
thunk, and ROCr. ROCr does these checks for us, and presents a
query that allows us to see if GWS entries are available for use.
If so, then we have all the lower-level technologies needed, and
we should enable cooperative groups support for HIP.
[ROCm/clr commit: 35a68596f9]
The maxSharedMemoryPerMultiProcessor attribute is meant to describe
the number of bytes of shared memory (LDS space in AMD terminology)
in each SM (CU in AMD terminology). For instance, on AMD GPUs this
is often 64KB per CU, and some Nvidia GPUs it's 96KB per SM.
This shared memory is a different address space from the normal
global memory. However, the current HIP-HCC properties fill this
in with a size that matches the totalGlboalMem property. This gives
a drastically too-high calculation for the amount of LDS space that
each CU has -- tens of GBs vs. 10s of KBs.
This patch fixes this by pulling the maxSharedMemoryPerMultiProcessor
property from the HSA pool that describes how much workgroup-local
space is available on each CU. The HSA runtime eventually pulls
this from the topology information about LDSSizeInKB, defined as
"Size of Local Data Store in Kilobytes per SIMD".
Previously, this HSA query was used to fill in the value of the
sharedMemPerBlock property. On today's AMD GPUs, we know that
the amount of LDS avaialble to the workgroup is identical to the
amount of LDS space in the CU. However, in the future this may
differ. As such, this patch changes around the order and fills
in the "PerMultiProcessor" property from the HSA query (since
what's what the query is defined to return), and then separately
fills in the "PerBlock" property as we know it.
[ROCm/clr commit: 44b16df84f]
Fix two errors in hipOccupancyMaxActiveBlocksPerMultiprocessor.
1) Fix a possible segfault if the user passed in a null pointer for
the numBlocks value.
2) Handle the situation when the user is asking for a block size
that is larger than what the target device can hold within a
single block.
[ROCm/clr commit: 6ae1b1a321]
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.
[ROCm/clr commit: 730f23829e]
* 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 c1b707ff63.
* Revert "Try to initialize the primary context on cuda"
This reverts commit d9cbd5a143.
* updated test for nvidia path
* Add c++11 option for nvcc
Co-authored-by: satyanveshd <53337087+satyanveshd@users.noreply.github.com>
[ROCm/clr commit: e085417370]
hipDeviceEnablePeerAccess returns success and adds peer into the list even if it is not accessible which creates problem in hipMalloc when it tries to share the ptr to peer device.
Proposed change is to check the access status before updating the peer list and update only when it can access the peer.
[ROCm/clr commit: e334d3d6ec]
* Device texture functions should not normalize the sampled pixel. This is already done by HW.
* Add support to use h/w capability for normalized float data convertion for driver API's
Co-authored-by: ansurya <50609411+ansurya@users.noreply.github.com>
[ROCm/clr commit: fa363ae721]
* Fix bug in LaunchKernel test
Instead of passing the address of the gpu buffer, pass the address
of the pointer that holds the address of the gpu buffer
* Fix hipLaunchKernel's kernarg buffer construction.
The hipLaunchKernel implementation should rely on ihipModuleLaunchKernel
to construct the kernarg buffer correctly based on kernel metadata.
* Fix a bug in get_functions where the Kernel_descriptor wasn't constructed with the correct kernarg layout information.
* Fix a bug in kernarg layout parsing dealing with kernel without any arg
* teach ihipModuleLaunchKernel to handle kernel without any arg
* Add a more interesting test
[ROCm/clr commit: 14e235378f]
* Add missing texturePitchAlignment member to the hipDeviceProp_t struct.
* Add missing hipDeviceAttributeTexturePitchAlignment enumerator to the hipDeviceAttribute_t enum.
* Initialize texturePitchAlignment to 256. This works for gfx9+, but is technically overaligned in most cases for pre-gfx9.
* Add the texturePitchAlignment property to the NVCC path.
[ROCm/clr commit: 6da0c82b78]
The current implementation skips this procedure for a given device
object when a global symbol is found in the cache. This is incorrect:
- There could be other undefined globals that have not been previously
encountered further down the list
- If a symbol is found in the cache, it doesn't need to be pinned again
but it still need to be defined for the current executable
Added special case for the printf buffer symbol (already pinned by HCC)
The bug was exposed by running printf on different GPUs.
[ROCm/clr commit: 8fc7cad90f]
=> New ROCr calculates pitch as per HSA specification and addrlib is used to check whether HW can support that configuration. Hence few texture tests are failing with HSA_EXT_STATUS_ERROR_IMAGE_PITCH_UNSUPPORTED.
=> Determine pitch for linear images and always pass rowpitch to HSA API's.
[ROCm/clr commit: 688b6e6af0]
Fixes SWDEV-218626 and SWDEV-218629
Changes:
- Revert "`static inline` in a header, just like excess sugar in a diet, causes bloat (#1692)"
This reverts commit 47725baf8c.
- Revert "Fix rocFFT build failure (#1777)"
This reverts commit f7bac3e057.
[ROCm/clr commit: c091635f5a]
* Detect when an explicit printf buffer flush is required
in a device/stream synchronization function.
* hip_module.cpp: add missing hc_am.hpp header
[ROCm/clr commit: 26b50e1e1b]
Siu Chi Chan
Sarbojit2019
Joseph Greathouse
Jeff Daily
Aryan Salmanpour
jglaser
zhaozhangjian
Evgeny Mankov
srinivamd
saleelk
Rahul Garg
Alex Voicu
Maneesh Gupta
ansurya
Paul
vsytch
satyanveshd
Jatin Chaudhary