Transitioned dpctl to require DPC++ 2023, removed host related functions, fixed linker crash, enabled SyclKernel.max_sub_group_size property #1028
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The issue manifested itself as "Relocation trucated to fit" error, and was caused by the large size of device code produced in debug build. The suggested solution is to use link option `-fsycl-link-huge-device-code`. See https://github.com/intel/llvm/blob/sycl/sycl/doc/UsersManual.md#link-options ``` FAILED: dpctl/tensor/_tensor_impl.cpython-39-x86_64-linux-gnu.so : && /opt/intel/oneapi/compiler/2023.0.0/linux/bin/icpx -fPIC -fsycl -O3 -Wall -Wextra -Winit-self -Wunused-function -Wuninitialized -Wmissing-declarations -fdiagnostics-color=auto -fstack-protector -fstack-protector-all -fpic -fPIC -D_FORTIFY_SOURCE=2 -Wformat -Wformat-security -fno-strict-overflow -fno-delete-null-pointer-checks -fsycl -g -Wall -Wextra -Winit-self -Wunused-function -Wuninitialized -Wmissing-declarations -fdiagnostics-color=auto -fstack-protector -fstack-protector-all -fpic -fPIC -D_FORTIFY_SOURCE=2 -Wformat -Wformat-security -fno-strict-overflow -fno-delete-null-pointer-checks -fsycl -O0 -ggdb3 -DDEBUG -fsycl-device-code-split=per_kernel -shared -o dpctl/tensor/_tensor_impl.cpython-39-x86_64-linux-gnu.so dpctl/tensor/CMakeFiles/_tensor_impl.dir/libtensor/source/tensor_py.cpp.o dpctl/tensor/CMakeFiles/_tensor_impl.dir/libtensor/source/simplify_iteration_space.cpp.o dpctl/tensor/CMakeFiles/_tensor_impl.dir/libtensor/source/copy_and_cast_usm_to_usm.cpp.o dpctl/tensor/CMakeFiles/_tensor_impl.dir/libtensor/source/copy_numpy_ndarray_into_usm_ndarray.cpp.o dpctl/tensor/CMakeFiles/_tensor_impl.dir/libtensor/source/copy_for_reshape.cpp.o dpctl/tensor/CMakeFiles/_tensor_impl.dir/libtensor/source/linear_sequences.cpp.o dpctl/tensor/CMakeFiles/_tensor_impl.dir/libtensor/source/eye_ctor.cpp.o dpctl/tensor/CMakeFiles/_tensor_impl.dir/libtensor/source/full_ctor.cpp.o dpctl/tensor/CMakeFiles/_tensor_impl.dir/libtensor/source/triul_ctor.cpp.o dpctl/tensor/CMakeFiles/_tensor_impl.dir/libtensor/source/device_support_queries.cpp.o -Wl,-rpath,::::::: && : /lib/x86_64-linux-gnu/crti.o: in function `_init': (.init+0xb): relocation truncated to fit: R_X86_64_REX_GOTPCRELX against undefined symbol `__gmon_start__' /tmp/icpx-a02203/_tensor_impl-411e70.o: in function `sycl.descriptor_reg': offload.wrapper.object:(.text.startup+0x4): relocation truncated to fit: R_X86_64_PC32 against `.data.rel.ro' /tmp/icpx-a02203/_tensor_impl-411e70.o: in function `sycl.descriptor_unreg': offload.wrapper.object:(.text.startup+0x14): relocation truncated to fit: R_X86_64_PC32 against `.data.rel.ro' /tmp/icpx-a02203/_tensor_impl-411e70.o:(.eh_frame+0x20): relocation truncated to fit: R_X86_64_PC32 against `.text.startup' /tmp/icpx-a02203/_tensor_impl-411e70.o:(.eh_frame+0x38): relocation truncated to fit: R_X86_64_PC32 against `.text.startup' /usr/lib/gcc/x86_64-linux-gnu/9/crtbeginS.o: in function `deregister_tm_clones': crtstuff.c:(.text+0x3): relocation truncated to fit: R_X86_64_PC32 against `.tm_clone_table' crtstuff.c:(.text+0xa): relocation truncated to fit: R_X86_64_PC32 against symbol `__TMC_END__' defined in .data section in dpctl/tensor/_tensor_impl.cpython-39-x86_64-linux-gnu.so crtstuff.c:(.text+0x16): relocation truncated to fit: R_X86_64_REX_GOTPCRELX against undefined symbol `_ITM_deregisterTMCloneTable' /usr/lib/gcc/x86_64-linux-gnu/9/crtbeginS.o: in function `register_tm_clones': crtstuff.c:(.text+0x33): relocation truncated to fit: R_X86_64_PC32 against `.tm_clone_table' crtstuff.c:(.text+0x3a): relocation truncated to fit: R_X86_64_PC32 against symbol `__TMC_END__' defined in .data section in dpctl/tensor/_tensor_impl.cpython-39-x86_64-linux-gnu.so crtstuff.c:(.text+0x57): additional relocation overflows omitted from the output dpctl/tensor/_tensor_impl.cpython-39-x86_64-linux-gnu.so: PC-relative offset overflow in PLT entry for `_ZSt10_ConstructIN4sycl3_V15eventEJEEvPT_DpOT0_' icpx: error: linker command failed with exit code 1 (use -v to see invocation) ```
|
@mdtoguchi @sergey-semenov I applied the suggested solution (learned in CMPLRLLVM-39897). Are there alternative ways to address the problem? Would splitting the library into several smaller SO files work as well? |
|
Array API standard conformance tests for dpctl=0.14.1dev0=py310h76be34b_47 ran successfully. |
@oleksandr-pavlyk, breaking up in to separate |
|
View rendered docs @ https://intelpython.github.io/dpctl/pulls/1028/index.html |
DPC++ 2023 has introduced a regression where the aspect returns 1 even for devices without fp64 aspect.
… since host device has been removed from 2023 compiler
With 2022.2 a non-zero value was being returned for interoperability kernel compiler for GPU device, but it now returns zero with 2023.0 compiler.
Debug build was used previsouly to accelerate the build, but due to growth of device section, debug build takes longer than the release build.
oleksandr-pavlyk
force-pushed
the
fix-debug-build-linker-relocation-truncated-to-fit
branch
from
7b3e89f to
37a3fe7
Compare
Moved host operations done after kernel submission to improve chance of detecting non-complete status of kernel submission.
Host device has been removed from DPC++ compiler in 2023.0.0 Also removed support for host device type and for backend host.
Removed `DPCTLDevice_IsHost`, `DPCTLHostSelector_Create`, `DPCTLContext_IsHost` and use of is_host from Python API.
oleksandr-pavlyk
force-pushed
the
fix-debug-build-linker-relocation-truncated-to-fit
branch
from
37a3fe7 to
9aa909f
Compare
oleksandr-pavlyk
changed the title
oleksandr-pavlyk
changed the title
oleksandr-pavlyk
changed the title
|
Array API standard conformance tests for dpctl=0.14.1dev0=py310h76be34b_63 ran successfully. |
|
Array API standard conformance tests for dpctl=0.14.1dev0=py310h76be34b_68 ran successfully. |
oleksandr-pavlyk
force-pushed
the
fix-debug-build-linker-relocation-truncated-to-fit
branch
from
b5f0f10 to
4b0f484
Compare
|
Array API standard conformance tests for dpctl=0.14.1dev0=py310h76be34b_68 ran successfully. |
|
Array API standard conformance tests for dpctl=0.14.1dev0=py310h76be34b_69 ran successfully. |
|
Array API standard conformance tests for dpctl=0.14.1dev0=py310h76be34b_71 ran successfully. |
Comment on lines
+148
to
+149
| ASSERT_TRUE(add_private_mem_sz >= 0); | ||
| ASSERT_TRUE(axpy_private_mem_sz >= 0); |
There was a problem hiding this comment.
This is actually an improvement in DPC++ RT, not a regression. Neither of the kernels use private memory.
oleksandr-pavlyk
deleted the
fix-debug-build-linker-relocation-truncated-to-fit
branch
|
Deleted rendered PR docs from intelpython.github.com/dpctl, latest should be updated shortly. 🤞 |
|
Array API standard conformance tests for dpctl=0.14.1dev0=py310h76be34b_71 ran successfully. |
This was referenced Jan 16, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The crash manifested itself as "Relocation trucated to fit" error, and was caused by the large size of device code produced in debug build. The suggested solution is to use link option
-fsycl-link-huge-device-code.See https://github.com/intel/llvm/blob/sycl/sycl/doc/UsersManual.md#link-options
Removed
DPCTLDevice_IsHost,DPCTLContext_IsHost,DPCTLHostSelector_Create, and Python API:dpctl.select_host_device,dpctl.has_host_device,dpctl.SyclDevice.is_host, anddpctl.SyclDevice.has_aspect_host, as well ashostbackend andhost_devicedevice type.Also added support for enabled in DPC++ 2023
dpctl.SyclKernel.max_sub_group_size.Removed use of
__SYCL_COMPILER_2023_SWITCHOVERpreprocessor constant and introduced__SYCL_COMPILER_VERSION_REQUIREDconstant inConfig/dpctl_config.h. Introducedstatic_assertinlibsyclinterface/source/*.cppfiles that#include <CL/sycl.hpp>to ensure that the compiler meets the minimum required version.