-
Notifications
You must be signed in to change notification settings - Fork 738
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[SYCL] Fix Ambiguity in Overloaded Unary Minus Operator for bfloat16 #15393
Commits on Mar 26, 2024
-
[SYCL][FPGA] Allow tablegen handle mutually exclusive decl attrs (SYC…
…LIntelRegister, SYCLIntelMemory) This patch uses MutualExclusions tablegen support to allow us to remove a custom diagnostic checking codes with FPGA attributes: [[intel:fpga_register]] and [[intel::fpga_memory]]. No test is added as we alreday have an existing LIT test (SemaSYCL/local.cpp) that shows the behavior.
Configuration menu - View commit details
-
Copy full SHA for 2457447 - Browse repository at this point
Copy the full SHA 2457447View commit details -
Configuration menu - View commit details
-
Copy full SHA for 7b7a5f3 - Browse repository at this point
Copy the full SHA 7b7a5f3View commit details
Commits on Mar 28, 2024
-
Configuration menu - View commit details
-
Copy full SHA for e41aacc - Browse repository at this point
Copy the full SHA e41aaccView commit details
Commits on Jun 27, 2024
-
Configuration menu - View commit details
-
Copy full SHA for ceb0eab - Browse repository at this point
Copy the full SHA ceb0eabView commit details
Commits on Jun 28, 2024
-
Configuration menu - View commit details
-
Copy full SHA for 4d9dfa1 - Browse repository at this point
Copy the full SHA 4d9dfa1View commit details
Commits on Jul 3, 2024
-
Configuration menu - View commit details
-
Copy full SHA for c133d97 - Browse repository at this point
Copy the full SHA c133d97View commit details
Commits on Jul 13, 2024
-
Configuration menu - View commit details
-
Copy full SHA for 76b7e48 - Browse repository at this point
Copy the full SHA 76b7e48View commit details
Commits on Jul 16, 2024
-
Configuration menu - View commit details
-
Copy full SHA for feaa623 - Browse repository at this point
Copy the full SHA feaa623View commit details -
[SYCL] Fix group algorithms for non-uniform groups, marray and vec (i…
…ntel#14364) This commit makes the following fixes to group algorithms: * Some GroupNonUniform SPIR-V builtins were incorrectly named as they were bundled together with their KHR Group SPIR-V equivalents. These have been renamed to map correctly to the right SPIR-V operations. * `sycl::marray` is now considered when checking for arithmetic types, making it usable in group broadcast operations and the functions that use them. * The representation of `bool` in `sycl::vec` has been changed to unsigned to match the representation picked by `ConvertToOpenCLType`. * The representation of `char` in `sycl::vec` has been changed to to match the representation picked by `ConvertToOpenCLType` to avoid cases where signedness would cause type mismatches. --------- Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
Configuration menu - View commit details
-
Copy full SHA for d29de7e - Browse repository at this point
Copy the full SHA d29de7eView commit details -
[SYCL][NVPTX] Emit reqd_work_group_size attributes as NVVM annotations (
intel#14502) Only emit the provided values as annotations in the LLVM IR. The NVPTX backend will pad missing values with 1s. This suits the fact that the attribute must provide as many values as the dimensionality of the work-group, and we can assume that the work-group size of unused dimensions is 1.
Configuration menu - View commit details
-
Copy full SHA for de655a3 - Browse repository at this point
Copy the full SHA de655a3View commit details -
[SYCL][Test] Add test for ABI-neutrality of sycl classes (intel#14558)
It is difficult to understand how to detect all possible data members which can cross ABI boundaries and cause problems because of dual ABI issue. For now using this approach which covers most of the classes, more than currently covered by "sycl/test/abi/layout*" tests. Current hits are in sycl::detail CG classes and subclasses (MFileName, MFunctinName, MKernelName and PipeName string data members). Even though I'm not sure if those cross ABI boundaries or not, I think we have to be conservative and fix those too. Using -fdump-record-layouts-complete to get dump for all complete record types, see description in https://reviews.llvm.org/D104484 for details.
Configuration menu - View commit details
-
Copy full SHA for 4d1ef3a - Browse repository at this point
Copy the full SHA 4d1ef3aView commit details -
[SYCL][NFCI][ABI-Break] Move handler members to impl (intel#14460)
This moves some of the members in the handler class to its impl class. Doing so allows developers to change these arguments without breaking ABI. This also moves more implementation details, such as command-group classes, launch configuration information, argument information and HostTask tracking, into sources to avoid hard-to-find ABI breaks in the communication between headers and runtime library. In addition to this, the following improvements are made: * The HostKernel class has been simplified to no longer have call and runOnHost functions. * The HostTask wrapper class has been moved to sources and the owner has been changed from a unique_ptr to a shared_ptr, which prevents the need for including host_task_impl.hpp in odd places. --------- Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
Configuration menu - View commit details
-
Copy full SHA for 0c368b1 - Browse repository at this point
Copy the full SHA 0c368b1View commit details -
[CI] Make Windows LIT_OPTS match Linux (intel#14583)
The `LIT_OPTS` were different compared to Linux so the output logs from CI runs is different. I noticed this when searching for passed tests in the Windows log and noticed it wasn't there. Signed-off-by: Sarnie, Nick <nick.sarnie@intel.com>
Configuration menu - View commit details
-
Copy full SHA for 3657bd9 - Browse repository at this point
Copy the full SHA 3657bd9View commit details -
[SYCL][ABI-Break] Fold host_half_impl::half into half_impl::half (int…
…el#13597) This commit folds the implementation of host_half_impl::half into half_impl::half and making the vector element representation the same as the half representation. This allows us to avoid strict alias violation for half vectors in their operator[] implementations. Note that this is marked as an ABI break as it removes symbols on Windows, despite these symbols never being in the library. --------- Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
Configuration menu - View commit details
-
Copy full SHA for 137fd1f - Browse repository at this point
Copy the full SHA 137fd1fView commit details -
[SYCL][E2E] Fix two Windows ocloc tests (intel#14560)
I recently found out ocloc tests weren't running in Windows CI because the ocloc tool wasn't installed on the runners. When trying to fix it, I [hit](https://github.com/intel/llvm/actions/runs/9909706895/job/27379893885?pr=14114) two failures that would have always failed if we were testing (the other two are going to be fixed [here](intel#14556)). The first fix disables the test on Windows because it is using `pvc` which is not available on Windows. The seconds add a requirement for the OCL CPU driver to be installed because the test is using `ocloc-aot`, as per the [ocloc-aot doc](https://github.com/intel/llvm/blob/56e88d591c52a978abdd5e4279853311cae4a55e/opencl/opencl-aot/README.md?plain=1#L11). The CI testing in this PR is only confirming I didn't break the tests on non-win ocloc because I haven't enabled Windows ocloc testing yet, but I manually verified the fix. --------- Signed-off-by: Sarnie, Nick <nick.sarnie@intel.com>
Configuration menu - View commit details
-
Copy full SHA for 89a4e16 - Browse repository at this point
Copy the full SHA 89a4e16View commit details -
Configuration menu - View commit details
-
Copy full SHA for 019d080 - Browse repository at this point
Copy the full SHA 019d080View commit details -
[SYCL] Make swizzle mutating operators const friends (intel#13012)
In intel#12682 the mutating operators for swizzles (+=, -=, ..., ++, --) were reverted to be members rather than friends. Since swizzles mutate the underlying vec rather than themselves these operators should take and return constant references instead, which this commit implements. --------- Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
Configuration menu - View commit details
-
Copy full SHA for 8f8644c - Browse repository at this point
Copy the full SHA 8f8644cView commit details -
[SYCL] kernel_compiler include file paths collision fix (intel#14490)
kernel_compiler include_file support shouldn't have files that might collide. Spec has been recently clarified as well ( intel@a6d8758 )
Configuration menu - View commit details
-
Copy full SHA for 94a10a6 - Browse repository at this point
Copy the full SHA 94a10a6View commit details -
[NFC][SYCL] Update tests to reflect new offload model (intel#14539)
Update existing SYCL based driver tests to use the new offload model and retain the equivalent old offloading model tests be creating copies. These tests use the --offload-new-driver and --no-offload-new-driver options to force a particular model to follow.
Configuration menu - View commit details
-
Copy full SHA for 5401bac - Browse repository at this point
Copy the full SHA 5401bacView commit details -
Configuration menu - View commit details
-
Copy full SHA for ac20312 - Browse repository at this point
Copy the full SHA ac20312View commit details -
[FPGA] [CodeGen][NFC] Enhance Intel FPGA annotation function and opti…
…mize code (intel#14577) This commit makes several improvements to the `CodeGenModule::addGlobalIntelFPGAAnnotation` method: - Updating the comment style to be Doxygen compliant for the `addGlobalIntelFPGAAnnotation` function. - Replaced `getAs<RecordType>()` with `castAs<RecordType>()` after confirming that the variable declaration type is guaranteed to be a `RecordType`. This change removes the need for a null check and streamlines the code. - Modified the base class iteration loop to use a reference (`const auto &Base`) to avoid unnecessary copies of `CXXBaseSpecifier` objects. These changes enhance code readability, improve code documentation and potentially increase performance by avoiding unnecessary dynamic type checks and copies. --------- Signed-off-by: Soumi Manna <soumi.manna@intel.com>
Configuration menu - View commit details
-
Copy full SHA for c4fdc92 - Browse repository at this point
Copy the full SHA c4fdc92View commit details
Commits on Aug 15, 2024
-
Configuration menu - View commit details
-
Copy full SHA for 4148548 - Browse repository at this point
Copy the full SHA 4148548View commit details
Commits on Sep 13, 2024
-
Configuration menu - View commit details
-
Copy full SHA for 3856438 - Browse repository at this point
Copy the full SHA 3856438View commit details -
[SYCL] Fix Ambiguity in Overloaded Unary Minus Operator for bfloat16
Signed-off-by: Soumi Manna <soumi.manna@intel.com>
Configuration menu - View commit details
-
Copy full SHA for a78dc15 - Browse repository at this point
Copy the full SHA a78dc15View commit details -
Configuration menu - View commit details
-
Copy full SHA for 6a65b8f - Browse repository at this point
Copy the full SHA 6a65b8fView commit details -
Configuration menu - View commit details
-
Copy full SHA for 4802b39 - Browse repository at this point
Copy the full SHA 4802b39View commit details -
Configuration menu - View commit details
-
Copy full SHA for b9dea3e - Browse repository at this point
Copy the full SHA b9dea3eView commit details