LLVM and SPIRV-LLVM-Translator pulldown (WW25) #10311
Closed
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
Includes a large number of missing `pi_result` mappings, so previously reported errors should no longer map to just `PI_ERROR_UNKNOWN`. NFCI for the adapters.
We are going to re-build the cuda runner, need to ensure CI continues working while it will be offline. The move is also good by itself as there was nothing cuda-specific in those tasks.
We've switched the task to use "[Linux, build]" runners in #10074 and this option isn't needed anymore.
…sent (#10079) Aligns the task with its Linux counterpart.
Due to issue #7268 I changed old exceptions to new ones (`sycl::exception`)
`std::true_t` and `std::false_t` don't exist, the correct ones are `std::true_type` and `std::false_type`. Signed-off-by: Maronas, Marcos <marcos.maronas@intel.com>
#9546 and similar accidentally broke code that was responsible to reset Intel GPU before running End-to-End tests. Restore functionality + add reset for the pre-commit task running End-to-End tests with a nightly build.
Signed-off-by: jinge90 <ge.jin@intel.com>
This patch updates GPU RT version on Linux to support [DEVICE_IP_VERSION_EXT](oneapi-src/level-zero-spec#23) implementation required for #9843.
Certain quotient values are off by 1 since the conversion to `int` truncates the decimal value. Use `std::round` to round to the nearest integer before converting to `int`. Signed-off-by: Michael Aziz <michael.aziz@intel.com>
These checks are causing issues for very large USM allocations because the `MAX_MEM_ALLOC_SIZE` reported is lower than what CUDA actually supports. We will follow up with an update on the reported `MAX_MEM_ALLOC_SIZE`, but it makes sense to remove the checks either way, as the CUDA allocation functions will return an error if they can't allocate the memory.
These tests now pass after this commit: 8417687 > Unexpectedly Passed Tests (2): SYCL :: InvokeSimd/Feature/ImplicitSubgroup/SPMD_invoke_ESIMD_external.cpp SYCL :: InvokeSimd/Feature/SPMD_invoke_ESIMD_external.cpp
…s images extension (#9808) This PR stands as the first PR of four to make code review easier, covering the changes made to libclc. See the original whole [PR](#9665). This is the first part of the implementation of revision 3 of the [bindless images extension proposal](https://github.com/intel/llvm/blob/8a48106617a9e2365c1935302d20e53bc7174368/sycl/doc/extensions/experimental/sycl_ext_oneapi_bindless_images/sycl_ext_oneapi_bindless_images.asciidoc). --------- Co-authored-by: SYCL Unbound Team <sycl.unbound@codeplay.com> Co-authored-by: Isaac Ault <isaac.ault@codeplay.com> Co-authored-by: Hugh Bird <hugh.bird@codeplay.com> Co-authored-by: Duncan Brawley <duncan.brawley@codeplay.com> Co-authored-by: Przemek Malon <przemek.malon@codeplay.com> Co-authored-by: Chedy Najjar <chedy.najjar@codeplay.com> Co-authored-by: Sean Stirling <sean.stirling@codeplay.com> Co-authored-by: Peter Zuzek <peter@codeplay.com>
) This change implements `sycl_ext_intel_grf_size`, and in particular: 1) Moves the `sycl_ext_intel_grf_size` spec document from the `proposed` folder to the `experimental` folder, and updates the implementation status in the document to match. 2) Adds two new kernel properties `sycl::ext::intel::experimental::grf_size` and `sycl::ext::intel::experimental::grf_size_automatic`, as per the spec. `grf_size` adds the `sycl-grf-size` metadata with a value of the template parameter (`128` or `256`). `grf_size_automatic` adds the `sycl-grf-size` metadata with a value of `0`. 3) Marks the `sycl::detail::register_alloc_mode` property as deprecated, and it still works as before. 4) Updates `CompileTimePropertiesPass.cpp` to map the `sycl-grf-size` metadata added by the front-end to the `RegisterAllocMode` metadata which `llvm-spirv` looks for. This `RegisterAllocMode` metadata is how AOT works. 5) Updates `sycl-post-link` to split by the `sycl-grf-size` metadata, have a `sycl-grf-size` binary property, and do a error check to make sure the deprecated `sycl::detail::register_alloc_mode` and `grf_size`/`grf_size_automatic` are not set at the same time on the same kernel. 6) Updates `program_manager` to deal with the new image property and pass the right flags 7) Updates exists tests to also test the new properties. --------- Signed-off-by: Sarnie, Nick <nick.sarnie@intel.com>
…h flag to them (#10090) Signed-off-by: Tikhomirova, Kseniya <kseniya.tikhomirova@intel.com>
Signed-off-by: Tikhomirova, Kseniya <kseniya.tikhomirova@intel.com>
This test used to be only run on the CPU and GPU but when we moved to
the `%{run}` notation, it wasn't marked as unsupported on accelerator.
It seems unsupported on accelerator. The `clCreateProgramWithIL()` call
is returning and error -59 when the accelerator device is the one behind
the context. I went back and checked and it has never worked on
accelerator. I'll make a note in the online compiler tickets and if this
is a real error, a ticket will be opened. But the test should be
restored so to not be run on accelerator.
…ml (#10100) Two reasons: 1) Not reliable, easy to skip testing without noticing. 2) That accidentally enabled "opencl:acc" that we don't run anywhere else and its state isn't green, resulting in "false" failures reported by CI.
…10076) It is needed to handle suite's run_launcher param properly.
In the continued effort to make the math builtin functions into overloads, this commit changes the scalar and vector relational builtins to adhere to the new builtin overload structure. --------- Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
Add support for -ftarget-export-symbols. This option when used with -fsycl -fsycl-targets=spir64_gen when creating shared libraries. This exposes the exported symbols in the library so they can be available for other modules.
Introduces support for work-group local allocations at program and function scope, lifting some of the restrictions of sycl_ext_oneapi_group_local_memory. --------- Signed-off-by: John Pennycook <john.pennycook@intel.com> Co-authored-by: Greg Lueck <gregory.m.lueck@intel.com>
…rk_group_size (#10093) According to SYCL 2020, > If a kernel is decorated with this attribute and then submitted to a device that does not support the work group size, the implementation must throw a synchronous exception with the `errc::kernel_not_supported` error code. This patch fixes sycl::errc in such case: it returned `errc::nd_range`, now it returns `errc::kernel_not_supported`.
…iver update (#10111) Fixes post-commit: #10054 (comment)
have @intel/dpcpp-nativecpu-pi-reviewers own: sycl/plugins/native_cpu/ sycl/test/check_device_code/native_cpu/ sycl/test/native_cpu/ clang/include/clang/Basic/SYCLNativeCPUHelpers.h clang/test/CodeGenSYCL/native_cpu_basic.cpp clang/test/Driver/sycl-native-cpu-fsycl.cpp clang/test/Driver/sycl-native-cpu-warn.cpp clang/test/Driver/sycl-native-cpu.cpp llvm/include/llvm/SYCLLowerIR/EmitSYCLNativeCPUHeader.h llvm/include/llvm/SYCLLowerIR/PrepareSYCLNativeCPU.h llvm/lib/SYCLLowerIR/EmitSYCLNativeCPUHeader.cpp llvm/lib/SYCLLowerIR/PrepareSYCLNativeCPU.cpp sycl/doc/design/SYCLNativeCPU.md --------- Co-authored-by: PietroGhg <38155419+PietroGhg@users.noreply.github.com> Co-authored-by: Alexey Bader <alexey.bader@intel.com>
I plan to add automatic "lastest-good-nightly" release to it, we need to extend the testing for that.
have @intel/syclcompat-lib-reviewers own: sycl/include/syclcompat/ sycl/doc/syclcompat/ sycl/unittests/syclcompat/ sycl/cmake/modules/AddSYCLLibraryUnitTest.cmake sycl/include/syclcompat.hpp which equates to sycl/**/syclcompat/ @intel/syclcompat-lib-reviewers sycl/cmake/modules/AddSYCLLibraryUnitTest.cmake @intel/syclcompat-lib-reviewers sycl/include/syclcompat.hpp @intel/syclcompat-lib-reviewers
Added to the spec in this PR: KhronosGroup/SYCL-Docs#351 Fixes #9786
…10252) Signed-off-by: jaime.a.arteaga.molina <jarteaga@scsel-cfl-10.sc.intel.com> Co-authored-by: jaime.a.arteaga.molina <jarteaga@scsel-cfl-10.sc.intel.com>
…ers (#9914) Remove the thread local variable workaround for deferred release of auxiliary resources by passing the required information from reduction headers.
It was only applied to definitions.
Same as #9419. This updates a few tests that were missed where the `fast-math` flag affects at least the cuda backend. These tests assume `no-fast-math` precision. --------- Signed-off-by: JackAKirk <jack.kirk@codeplay.com> Co-authored-by: JackAKirk <chezjakirk@gmail.com>
This implements the current extension doc from #6104 in the CUDA backend only. Fixes #7543. Fixes #6749. --------- Signed-off-by: JackAKirk <jack.kirk@codeplay.com> Co-authored-by: Nicolas Miller <nicolas.miller@codeplay.com> Co-authored-by: JackAKirk <chezjakirk@gmail.com> Co-authored-by: Steffen Larsen <steffen.larsen@intel.com>
Add noexcept to AccessorBaseHost::getPtr since get_pointer is noexcept in SYCL 2020
… with -fsycl (#9864) This PR allows users to include a Pre-Compiled header (PCH) file while performing the host compilation step with `-fsycl`. **// Create PCH** `clang++ -x c-header test.h` **// Use PCH** ``` clang++ -c -fsycl -include test.h test.cpp clang++ -c -fsycl -include-pch test.h.gch test.cpp ```
Update the description of the limitations in the "sycl_ext_oneapi_device_architecture" extension. The newly added host APIs in that extension do not have the same restrictions as the `if_architecture_is` function.
…L 1.2.1 (#10174) * This PR covers updates in [Default multi_ptr to legacy to avoid code break with SYCL 1.2.1](KhronosGroup/SYCL-Docs#432) * Declare access::decorated::legacy as deprecated. * Revert `get_pointer` interface for device, host and local_accessor back to 1.2.1 version. * `enable_if` `async_work_group_copy` in `group.hpp` and `nd_item.hpp` to ensure Dst and Src(const / non-const) have the same type. * Add multi_ptr::get_raw and multi_ptr::get_decorated member functions to the access::decorated::legacy specialization of multi_ptr.
This patch combines ADL-S and RPL-S architectures for sycl_ext_oneapi_device_architecture extension. These two architectures are the same. Fixes #10184
The loading address didn't calculate the column offset right w.r.t the sub-group id. Signed-off-by: Yilong Guo <yilong.guo@intel.com>
Upcoming changes to SYCL builtins will cause code using namespace sycl implicitly to be ambiguous when using certain builtins that are available in the global scope already. This commit changes uses of `abs` to be namespace qualified to avoid them breaking with future builtin changes. --------- Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
The workflow will be shared between normal "Build + E2E" and "E2E on Nightly" tasks.
This change better aligns checkout between build/e2e-tests. Next step would be to make them use the same commit for the merge but I plan to do that in a separate PR.
Chenyang-L
added
the
disable-lint
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.
LLVM: llvm/llvm-project@af35be5
SPIRV-LLVM-Translator: KhronosGroup/SPIRV-LLVM-Translator@9823690
Community has started to refactor some code assuming opaque pointer only,
however the SPIRV component used in sycl is not ready to enable the opaque pointer by default,
so we have to revert some of the commit by guarding them within macro INTEL_SYCL_OPAQUEPOINTER_READY,
and keep both version of the code.
The old code should be removed once we enable opaque pointer by default.