Minimum queue size was not enforced at the Agent level. Minimum
size should be one page to give unifority across all asics.
Change-Id: I26394f79458d09fbceb79fc8aaf495e2c26a8ff3
On gfx90a only a reduced number of CUs must be used for cooperative
dispatches due to CWSR and launcher interactions with asymetric
harvest. We must use one fewer CUs per SE than the lowest count of
CUs on any SE.
Also adds env var HSA_COOP_CU_COUNT which enables the cooperative
CU count computation. Set to 1 to enable the new computation.
This is an opt-in feature that will become enabled by default (opt-out)
in a future release.
Change-Id: Ifbb75ced3bbc15876eef44922c6a4f6fde8c4c28
Corrections have been made in libhsakmt, and corresponding changes are required here as well.
Signed-off-by: Chen Gong <curry.gong@amd.com>
Change-Id: Ib697ce25278c2c5ac6ef0206930ec285f46c60d1
Incease more timeout according to peers number to pass the
test on some PCIe link platforms.
Change-Id: Ifcb8c7297d6960c96fc18d29bc0a48733ca50165
Signed-off-by: Eric Huang <jinhuieric.huang@amd.com>
The start iterator becomes invalid after it is removed from
std::map prefetch_map_. This was causing a segfault when the iterator is
incremented afterwards.
Signed-off-by: David Yat Sin <david.yatsin@amd.com>
Change-Id: I4b0b763d2cb4ee99c0b8571c2c526b834e74077a
Mapped memory areas become invalid after fork, and the child process is
required to remap the memory areas after a fork. So we mark these device
memory mappings with MADV_DONTFORK so that they are removed from the
child process after fork.
This was causing some issues when doing CRIU checkpoint/restore because
CRIU and amdgpu_plugin were not able to handle these mappings.
Change-Id: I50eb334aecea6dab7522d94da0273adcf4fb1ce0
Signed-off-by: David Yat Sin <david.yatsin@amd.com>
Total VRAM size on APU is 512M usually,
Framebuffer also is allocated from VRAM.
There is no enough memory for this case.
/home/ruiliji2/p5/libhsakmt/tests/kfdtest/src/KFDMemoryTest.cpp:1285: Failure
Value of: (hsaKmtMapMemoryToGPUNodes(bufs[i], bufSize, &altVa, mapFlags, 1, &defaultGPUNode))
[ FAILED ] KFDMemoryTest.MMBench (1034 ms)
Change-Id: Ib4201291122d85f6512a85859aea9a4713fb4f5c
(cherry picked from commit a9f924484e7022a2d53ee02811b080f0833eba55)
skip HDP flush test when remap feature is not supported.
Backgroud:
the HDP register remap is skipped in sriov mode,
it will cause mmio base is nullPtr.
Signed-off-by: Yang Wang <KevinYang.Wang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Change-Id: Ib9aea1900931e30571656397a485ee4db051ec0a
Explicitly free the user buffer ptr before test's tear down. Otherwise
the svm_bo object will never be released, causing a BUG error. Due to
a late callback to svm_migrate_page_free when prange not longer exist.
Also did cosmetic adjustments.
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
Change-Id: I989c62de8a9634faa84e42def956cecb3f84e329
The AMD compiler team has confirmed that they expect gfx90c
to be gfx90c, with a major/minor/stepping of 9, 0, and 12
respectively. It appears that there is a typo in the libhsakmt
topology information that lists this part as gfx902. This patch
fixes the issue.
Signed-off-by: Joseph Greathouse <Joseph.Greathouse@amd.com>
Change-Id: I6f907a7aa6f190b12aba8bb4210c7b341b3c720b
This is causing issues with side by side, sorry for the noise.
This license location isn't ideal but it's good enough for now.
Change-Id: Iba2a84cedf22466fdaaf3c63b6ea49c9fc277967
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
Calling cmake replaces this file, so no need to commit it.
Change-Id: Ic4747cc9eebd9cbfc61d524a31d2025c04eda12e
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
The copyright file will conflict if multiple thunks are installed. This
should resolve the issue by adding the version to the install path.
Change-Id: Ieac5a3eba979b3e934fb9100f890b92fc7c35d71
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
Prior solution used a single global lock to protect the memory tracking structures.
This change protects the memory tracking structure with a shared mutex (rw lock) in
shared (r) mode for memory allocations and frees so that long duration processes,
calling to kfd, can be done in parallel. Operations which must modify the memory map
take the mutex in exclusive mode (w) and must not call to the thunk while holding
the mutex.
The fragment allocator now requires separate protection and is protected with a
mutex at the device level. Protecting at the device level, rather than pool,
allows retention of the current recursive design and allows calling Trim from
withing Allocate. This could be made finer (pool level locks) but would
require backing out of Allocate entirely to call Trim. Trim and any retried
Allocation must be done in isolation (per device) or we may report OOM when
memory is actually available in some pool's fragment cache. So some device
level serialization is required in at least some paths.
Change-Id: I7c1e94d6965ffcc602b12fefdd3a6e97b84b5e00
CPACK doesn't have proper logic for installing the license as described
by CPACK_RESOURCE_FILE_LICENSE.
For Debian packaging, the license is expected to be installed as:
/usr/share/doc/PACKAGENAME/copyright
To do this, I've added a bit of logic for CPACK to copy this into the
package using CPACK_INSTALL_COMMANDS to prep the directory, and
CPACK_INSTALLED_DIRECTORIES to add it to the package. This applies to
both RPM and DEB, so I've added some logic to the spec file to exclude
this file (note that CPACK_RPM_EXCLUDE_FROM_AUTO_FILELIST_ADDITION does
not work for files installed with CPACK_INSTALLED_DIRECTORIES).
For RPM install, I've just added a small bit of logic to the spec file
to handle it. The file needs to be copied into the spec working
directory, then a macro is used to handle the rest. Note the license
macro does not work on EL6, but I don't think we want to support this.
Change-Id: I06ce63d300419893cb8274bc504a15633e304d91
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
I used the binary directory instead of the source directory to specify
the spec.in path, which passed local testing since these directories
are in the same location. This is not guarenteed to be true.
Change-Id: I1b49ca8453b9c074a947104c26fb39667d728a8e
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
CPack does not support recommends for RPM generation, so I've generated
a template RPM SPEC files in order to make modifications to allow for
support of recommends.
The spec.in file was generated using the cpack option
"CPACK_RPM_GENERATE_USER_BINARY_SPECFILE_TEMPLATE" and was modified very
sparingly to avoid any maintanance burden, e.g. can be easily
regenerated. The CPACK_RPM_USER_BINARY_SPECFILE is then used to specify
the customized template file, instead of using the cmake's template.
From what I understand, the point of these two options is to allow
developers to tailor the specfile to their desire, since rpm spec files
are much more advanced then the equilivent debian file.
Change-Id: I80c69be58a3c57729ed997fd2ce01f5d16b9e9b9
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
For the use of libdrm-amdgpu-amdgpu1 and libdrm-amdgpu, we should use
recommends, as we want these packages installed with a strong dependency
but avoid a strict dependency, since this is enhancement feature.
Using the newer libdrm, which is build for amdgpu-dkms, is ideal since
it will produce more correct marketing names, but should not be mandated
due to two reasons:
- A user may not want to install both libdrms on their system
- The system might not have the newer libdrm available
This patch only fixes the Ubuntu/debian package since recommends is not
properly implemented for the RPM generator for CPACK. For now,
"suggests" will have to do, since it's the closest option we have. I
will investigate if we can get around this issue.
Change-Id: I33a90c3ead235bbbe265238c026933688ea63fe3
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
Comments call out the specific operation being selected since the
ternary nest is a bit hard to read.
Change-Id: If033dbaa6cba132e96196ad3fc6d5572042041f4
Argument must be checked for nullptr before being dereferenced and
filled with the default return value.
Change-Id: I9ff366f066a5e18c78129bf59cc3ba00fca3ef18
SUGGESTS doesn't get installed by default by dpkg, it requires
--install-suggests. While the libdrm-amdgpu-amdgpu1 lib might not be
critical to core thunk usage, it's necessary enough to increase the
dependency from weak to strong, thus we use RECOMMENDS instead
Since CPack doesn't support RECOMMENDS for RPMs, and SUGGESTS isn't
strong enough, we just set it as REQUIRED. This should handle any
standard installation situations instead of trying to be fancy to try to
make things exactly the way we want them to be.
Signed-off-by: Kent Russell <kent.russell@amd.com>
Change-Id: I5a676fb4a3fbf5c7b35f503acc30cf01c056266d
This really should be set to conform to distro standards.
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
Change-Id: I8c3bdcc7eb103cec9db6aa9f9cfec25754784be8
This really should be set to conform to distro standards.
Change-Id: I63442b9ad58f70c4691c499b666f7c359dc9609d
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
On gcc-10.3.0 environment, hsa-runtime building is failed as below log:
compute/hsa/runtime/rocrtst/suites/negative/queue_validation.cc:470:18: error: conversion from ‘unsigned int’ to ‘uint16_t’ {aka ‘short unsigned int’} changes value from ‘4294967295’ to ‘65535’ [-Werror=overflow]
470 | aql().header |= 0xFFFFFFFF << HSA_PACKET_HEADER_TYPE;
| ~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
cc1plus: all warnings being treated as errors
make[2]: *** [CMakeFiles/rocrtst64.dir/build.make:339: CMakeFiles/rocrtst64.dir/home/aaliu/work/compute/hsa/runtime/rocrtst/suites/negative/queue_validation.cc.o] Error 1
make[2]: *** Waiting for unfinished jobs....
Signed-off-by: Aaron Liu <aaron.liu@amd.com>
Change-Id: I95fe72030368abc211b4b97b5a7ba00b5e094730
The infra is in place to make sure that we can get around the issue
where older versions of rpm don't know what "Suggests" is. Using the
HSAKMT_REQUIRES_LIBDRM variable will force the package to require
libdrm-amdgpu1 instead of just suggesting it in newer distros.
Change-Id: I5d595c4e4be5ad8e8c411dcec3a58365e1984f05
We need to add some more information about the debug features supported
by the platform. We are adding the following:
- debug supported
- dispatch info always valid
- precise memop supported
- watchpoints shared
Change-Id: I68deed98619396d17e28c6e18bad424b58297485
Signed-off-by: Philip Cox <Philip.Cox@amd.com>
Since it's shader-based, the GFX9 shader won't work for GFX10. Blacklist
it for now until we move to a more ASIC-agnostic shader
Signed-off-by: Kent Russell <kent.russell@amd.com>
Change-Id: I353e9a3fe8c7ba6f33556e39c7e4f6b36e010480
GetGlobalMemoryPool had improper return codes for an iterator callback
and did not properly order the APU pool selection path.
Change-Id: I01ab9d23e2352be98d9718bc25889ad4f779d3ca
Clang warns about bitwise operators on bools. Cast to int silences
the warning without introducing short circut logic.
Change-Id: I6e25138e1acf4a5562d3925ea5b2fcef3addb783
Would be nice to get warning count changes highlighted in CI though.
Clang's increasingly suspect diagnostics has caused multiple build
breaks without highlighting any actual issues.
Also: https://embeddedartistry.com/blog/2017/05/22/werror-is-not-your-friend/
Change-Id: I7dc82da58cd86f7b4f1a9fb511c4c039419271d4
This is required for Marketing Name, but Marketing Name isn't a hard
requirement for ROCT, so make it a Suggested package for DEBs.
Adding a commented-out possible solution for RPM-based distros, as
unfortunately SLE <10 and CentOS/RHEL <8 don't know what "SUGGESTS"
means. Potentially we will add a check that the build environment
that determines if we require the REQUIRES field to be used.
Hopefully we can work out the RPM side quickly, but for now enable
SUGGESTS on DEB as a start
Signed-off-by: Kent Russell <kent.russell@amd.com>
Change-Id: Icea689a14c185964e7f47fb23ee432f9b35ccb03
Due to a CPACK bug the package needs to remove header file
symlinks. Cleanup is required for uninstall and upgrade
since each release installs to a different folder.
Change-Id: I5ec378b21e69235404781c7bce3c0203eb38eed1
KFD topology has been corrected and the defaults used by this
workaround are no longer true for all chips.
Change-Id: I0242d8077e9666ed1cf0dc3985244258ae5c0924
For APU asics, the default configuration size of video memory is
relatively small, plus the reserved region, ratio of max alloc size to
the pool size may below the expected value, so adjust it.
Change-Id: I798b44d9532aa6a381a1cc19faa5a46110bf0ad6
The shader is incompatible with gfx1030, so blacklist it there until we get
to some more ASIC-agnostic shader generation
Signed-off-by: Kent Russell <kent.russell@amd.com>
Change-Id: I71f362a80e6e8f87ecece6afa6b34a5f02bc56bf
The return code is just -1 if any error occurs. To detect debugger
unavailable we need to check the actual ioctl error code.
Change-Id: I8a294c754196aec916809497ec8e810da2f072b8
Signed-off-by: Sean Keely <Sean.Keely@amd.com>
Early exit if the range is found to be fine grain. Indeterminate
should only apply if the range is neither coarse nor fine.
Change-Id: I54133e14f4e8cfa53e2d612f6112cdcdb5a47dfa
Sean Keely
Chen Gong
Jeremy Newton
Eric Huang
David Yat Sin
Ruili Ji
Yang Wang
Alex Sierra
Joseph Greathouse
Kent Russell
Aaron Liu
Philip Cox
Freddy Paul
Xiaomeng Hou