Summary: | =sci-libs/rocBLAS-5.1.3: fatal error: error in backend: Cannot select: 0x55756dea2098: i64 = FrameIndex<0> (broken by sys-devel/clang:15's 0008-clang-Driver-More-flexible-rules-for-loading-default.patch?) | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | hanetzer |
Component: | Current packages | Assignee: | Gentoo Science Related Packages <sci> |
Status: | RESOLVED FIXED | ||
Severity: | normal | CC: | gentoo, llvm, mgorny, sam, xgreenlandforwyy |
Priority: | Normal | Keywords: | PullRequest |
Version: | unspecified | ||
Hardware: | All | ||
OS: | Linux | ||
See Also: |
https://github.com/gentoo/gentoo/pull/29188 https://github.com/gentoo/gentoo/pull/33845 |
||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Attachments: | build.log |
Description
hanetzer
2023-01-11 00:48:20 UTC
Created attachment 848143 [details]
build.log
I'd also like to note that there are newer versions available. After bodging some of the patches to fit the new versions (5.3.3), it came down to the same build failure. I get the same Is this error specific to gfx1010? I'll try reproducing it on gfx1031
> Trying to get rocm running on my system (amd ryzen 2700/amd rx 6700 xt)
Also, rx 6700 xt should be gfx1031. According to build log I guess you're using 5700 xt?
(In reply to Yiyang Wu from comment #4) > Is this error specific to gfx1010? I'll try reproducing it on gfx1031 > Confirmed on gfx1031 I have tried: llvm/clang-15.0.6 + rocm-device-libs/rocm-comgr/hip-5.3.3 : fail llvm/clang-15.0.6 + rocm-device-libs/rocm-comgr/hip-5.1.3 : fail llvm/clang-15.0.0_rc2 + rocm-device-libs/rocm-comgr/hip-5.1.3 : success So I guess something went wrong in llvm/clang-15 releases. Will do some further testing (In reply to Yiyang Wu from comment #4) > Is this error specific to gfx1010? I'll try reproducing it on gfx1031 > > > Trying to get rocm running on my system (amd ryzen 2700/amd rx 6700 xt) > > Also, rx 6700 xt should be gfx1031. According to build log I guess you're > using 5700 xt? Ah yeh, misspoke/mistyped. Actual model is a: ASRock Radeon RX 5700 XT Taichi X (In reply to Yiyang Wu from comment #6) > I have tried: > > llvm/clang-15.0.6 + rocm-device-libs/rocm-comgr/hip-5.3.3 : fail > llvm/clang-15.0.6 + rocm-device-libs/rocm-comgr/hip-5.1.3 : fail > llvm/clang-15.0.0_rc2 + rocm-device-libs/rocm-comgr/hip-5.1.3 : success > > So I guess something went wrong in llvm/clang-15 releases. Will do some > further testing Further bisecting shows the problem may come from the llvm-gentoo-patchset llvm/clang-15.0.1::gentoo + rocm-device-libs/rocm-comgr/hip-5.3.3 : success llvm/clang-15.0.2::gentoo + rocm-device-libs/rocm-comgr/hip-5.3.3 : fail llvm/clang-15.0.2::gentoo with llvm-gentoo-patchset-15.0.1 + rocm-device-libs/rocm-comgr/hip-5.3.3 : success (In reply to Yiyang Wu from comment #8) > > llvm/clang-15.0.1::gentoo + rocm-device-libs/rocm-comgr/hip-5.3.3 : success > llvm/clang-15.0.2::gentoo + rocm-device-libs/rocm-comgr/hip-5.3.3 : fail > llvm/clang-15.0.2::gentoo with llvm-gentoo-patchset-15.0.1 + > rocm-device-libs/rocm-comgr/hip-5.3.3 : success As I tested, 0008-clang-Driver-More-flexible-rules-for-loading-default.patch can cause this bug (In reply to Yiyang Wu from comment #9) > As I tested, 0008-clang-Driver-More-flexible-rules-for-loading-default.patch > can cause this bug CC the author of 0008-clang-Driver-More-flexible-rules-for-loading-default.patch Could you try commenting stuff out in /etc/clang to see if any specific rule is causing it vs. loading config files at all? (In reply to Michał Górny from comment #11) > Could you try commenting stuff out in /etc/clang to see if any specific rule > is causing it vs. loading config files at all? Commenting out `-fstack-protector-strong` in `/etc/clang/gentoo-hardened.cfg` can mitigate this bug (In reply to Yiyang Wu from comment #12) > (In reply to Michał Górny from comment #11) > > Could you try commenting stuff out in /etc/clang to see if any specific rule > > is causing it vs. loading config files at all? > > Commenting out `-fstack-protector-strong` in > `/etc/clang/gentoo-hardened.cfg` can mitigate this bug Does that mean before clang-Driver-More-flexible-rules-for-loading-default.patch those configs never get loaded, so this bug is hidden? According to https://github.com/ROCmSoftwarePlatform/Tensile/issues/1455#issuecomment-1007632229 https://github.com/ROCmSoftwarePlatform/rocBLAS/issues/437 `-fstack-protector-strong` can cause errors when compiling GPU kernels. The simplest solution would be disabling this flag in rocm.eclass The bug has been closed via the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=bb6386121107a5a6f108a429e6042a237e07e371 commit bb6386121107a5a6f108a429e6042a237e07e371 Author: Yiyang Wu <xgreenlandforwyy@gmail.com> AuthorDate: 2023-01-20 10:12:16 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2023-01-21 03:32:16 +0000 dev-util/hip: append -fno-stack-protector for hipcc Closes: https://bugs.gentoo.org/890377 Signed-off-by: Yiyang Wu <xgreenlandforwyy@gmail.com> Closes: https://github.com/gentoo/gentoo/pull/29188 Signed-off-by: Sam James <sam@gentoo.org> dev-util/hip/files/hip-5.1.3-fno-stack-protector.patch | 16 ++++++++++++++++ .../hip/{hip-5.1.3-r2.ebuild => hip-5.1.3-r3.ebuild} | 1 + dev-util/hip/{hip-5.3.3.ebuild => hip-5.3.3-r1.ebuild} | 1 + 3 files changed, 18 insertions(+) The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=6992e6ff39664c798459db3f93e048a06095adf6 commit 6992e6ff39664c798459db3f93e048a06095adf6 Author: Sv. Lockal <lockalsash@gmail.com> AuthorDate: 2023-10-18 18:37:09 +0000 Commit: Benda Xu <heroxbd@gentoo.org> CommitDate: 2023-11-19 02:18:15 +0000 dev-util/hipcc-5.7.1: fix compilation issues - Restore the patch for adding -fno-stack-protector - Better handling of clang path using llvm.eclass - Fix the device lib path parsed to clang - QA: remove files installed to wrong location Bug: https://bugs.gentoo.org/890377 Bug: https://github.com/gentoo/gentoo/pull/33400 Signed-off-by: Sv. Lockal <lockalsash@gmail.com> Signed-off-by: Yiyang Wu <xgreenlandforwyy@gmail.com> Signed-off-by: Benda Xu <heroxbd@gentoo.org> Closes: https://github.com/gentoo/gentoo/pull/33845 .../files/hipcc-5.7.1-fno-stack-protector.patch | 13 ++++++++++++ ...hipcc-5.7.1-r1.ebuild => hipcc-5.7.1-r2.ebuild} | 24 +++++++++++++++++++--- dev-util/hipcc/hipcc-5.7.1.ebuild | 21 ------------------- 3 files changed, 34 insertions(+), 24 deletions(-) |