[SYCL] Improve SYCL_DEVICE_ALLOWLIST#3826
Merged
againull merged 8 commits intoJun 3, 2021
Merged
Conversation
This patch adds more stability to SYCL_DEVICE_ALLOWLIST: 1. Introduce 3 new keys `BackendName`, `DeviceType` and `DeviceVendorId` which should be used instead of `DeviceName` and `PlatformName`. These 3 new keys are more stable. 2. Refactor the implementation of SYCL_DEVICE_ALLOWLIST to make it more stable, to fix std::bad_alloc crash, and to make the code testable 3. Add unit tests for parsing SYCL_DEVICE_ALLOWLIST value and for functionality which allows device to use or reject it.
Contributor
Author
|
Working on fixing Jenkins/Precommit. |
bader
reviewed
May 27, 2021
| "BackendName:level_zero,SomeUnsupportedKey:gpu"); | ||
| } catch (sycl::runtime_error const &e) { | ||
| EXPECT_EQ(e.what(), std::string("Unrecognized key in SYCL_DEVICE_ALLOWLIST " | ||
| "-30 (CL_INVALID_VALUE)")); |
Contributor
There was a problem hiding this comment.
Why "-30 (CL_INVALID_VALUE)"?
Looks like some error code is mapped to OpenCL specific enumeration. The reason is unclear.
Contributor
Author
There was a problem hiding this comment.
Yes, AFAIS, this is a common situation in RT right now: there is no RT-only error codes, so everywhere in the code, even for non-backend related things, PI_ error codes are used, which are casted to CL_ error codes.
Contributor
There was a problem hiding this comment.
This is very annoying bug, which we should fix sooner rather than later.
Please, make sure it's addressed separately.
Contributor
Author
There was a problem hiding this comment.
Will create a tracker. @vladimirlaz, as you work closely to SYCL2020 exceptions, do you see this issue be reproducible? (showing PI_*** or CL_*** after throwing any exception)
Contributor
Author
|
Failures in Jenkins/Precommit should gone after intel/llvm-test-suite#304 |
Contributor
Author
bader
previously approved these changes
Jun 1, 2021
againull
reviewed
Jun 2, 2021
againull
self-requested a review
June 2, 2021 08:47
againull
reviewed
Jun 2, 2021
Contributor
|
Overall patch looks good to me, just several minor comments |
againull
previously approved these changes
Jun 2, 2021
bader
reviewed
Jun 3, 2021
againull
self-requested a review
June 3, 2021 16:04
againull
approved these changes
Jun 3, 2021
bader
approved these changes
Jun 3, 2021
Contributor
|
@dm-vodopyanov, please, fix post-commit fails in "no-assertions" configuration. |
alexbatashev
pushed a commit
to alexbatashev/llvm
that referenced
this pull request
Jun 4, 2021
* sycl: (320 commits) [SYCL] Silence a "local variable is initialized but not referenced" warning; NFC (intel#3870) [SYCL] Improve SYCL_DEVICE_ALLOWLIST (intel#3826) [SPIR-V] Change return value of mapType function (intel#3871) [SYCL] Fix post-commit failure in handler.hpp from unused-parameters. (intel#3874) [Driver][SYCL] Do not imply defaultlib msvcrt for Linux based driver on Windows (intel#3827) [SYCL] Unique stable name rebase (intel#3835) [SYCL] Align behavior of empty command groups with SYCL2020 (intel#3822) [SYCL][ESIMD] Make typenames and constants consistent with SYCL API style. (intel#3850) [SYCL] Allow __failed_assertion to support libstdc++-11 (intel#3774) [SYCL] Refactor stream class handing implementation (intel#3646) [SYCL] Fix syntax error introduced in intel#3401 (intel#3861) [SYCL] SYCL 2020 sub_group algorithms (intel#3786) [Buildbot][NFC] Add option to use LLD as linker (intel#3866) Revert "Emit correct location lists with basic block sections." [SPIRITTAnnotations] Fix debug info for ITT calls. (intel#3829) [SYCL][Doc] Fix build of Sphinx docs (intel#3863) [SYCL][FPGA][NFC] Tidy up intel_fpga_reg codegen test (intel#3810) [CODEOWNERS] Fix SPIRITTAnnnotations tests ownership (intel#3859) [SYCL][ESIMD] Host-compile simd.cpp test, fix errors & warnings. (intel#3846) [SYCL] Store pointers to memory allocations instead of iterators (intel#3860) ...
dm-vodopyanov
added a commit
to dm-vodopyanov/llvm
that referenced
this pull request
Jun 4, 2021
iclsrc
pushed a commit
that referenced
this pull request
Jul 17, 2026
…3826) Deduce storage class memory semantic from the pointer accessed by atomic intrinsics Inspired by llvm/llvm-project#193696, llvm/llvm-project#195049 Original commit: KhronosGroup/SPIRV-LLVM-Translator@0a6abf0df14b2f0
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
This patch adds more stability to SYCL_DEVICE_ALLOWLIST:
Introduce 3 new keys
BackendName,DeviceTypeandDeviceVendorIdwhich should be used instead ofDeviceNameandPlatformName. These 3 new keys are more stable.Refactor the implementation of SYCL_DEVICE_ALLOWLIST to
std::bad_alloccrashAdd unit tests for parsing SYCL_DEVICE_ALLOWLIST value and for
functionality which allows device to be accepted or to be rejected.