[SYCL-PTX] Add warp-reduce path in sub-group reduce #3949
Merged
Conversation
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
PTX introduces new warp reduction instructions with sm_80. These changes adds a path in select libclc subgroup collective functions for using these warp reduction instructions when available. As an effect of these changes, future changes to libclc targetting nvptx can use __nvvm_reflect to differentiate between SM versions, allowing architecture-specific paths that will be determined after linking device code with libclc. Signed-off-by: Steffen Larsen <steffen.larsen@codeplay.com>
Signed-off-by: Steffen Larsen <steffen.larsen@codeplay.com>
bader
approved these changes
Jun 21, 2021
|
/summary:run |
alexbatashev
added a commit
to alexbatashev/llvm
that referenced
this pull request
Jun 24, 2021
* upstream/sycl: (489 commits) [SYCL][NFC] Lower overhead of making plugin calls (intel#3982) [SYCL][NFC] Use default macro initialization where applicable (intel#3979) [SYCL] Enable SPV_INTEL_fpga_invocation_pipelining_attributes extension (intel#3864) [SYCL] Disable reassociate pass to reduce register pressure (intel#3615) [Driver][SYCL][FPGA] Restrict -O0 for FPGA with hardware (intel#3966) [SYCL][NFC] Fix warnings coming out of SYCL headers. (intel#3978) [SYCL] Fix bugs with recursion in SYCL kernel (intel#3958) [SYCL][LevelZero] Add support to detect host->device and device->host transfers for USM (intel#3975) [SYCL] Enable native FP atomics by default (intel#3869) [sycl-post-link] Avoid copying from nullptr (intel#3963) [SYCL-PTX] Add warp-reduce path in sub-group reduce (intel#3949) [BuildBot] Uplift CPU/FPGAEMU RT version for CI Process (intel#3946) Fix handling of complex constant expressions Handle OpSpecConstantOp with CompositeExtract and CompositeInsert Handle OpSpecConstantOp with VectorShuffle [FuncSpec] Don't specialise functions with NoDuplicate instructions. [mlir][linalg] Support low padding in subtensor(pad_tensor) lowering [gn build] Port 208332d [AMDGPU] Add Optimize VGPR LiveRange Pass. [mlir][Linalg] NFC - Drop unused variable definition. ...
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.
PTX introduces new warp reduction instructions with sm_80. These changes adds a path in select libclc subgroup collective functions for using these warp reduction instructions when available.
As an effect of these changes, future changes to libclc targeting nvptx can use
__nvvm_reflectto differentiate between SM versions, allowing architecture-specific paths that will be determined after linking device code with libclc.