45 Pull requests merged by 11 people

• Update .gitignore to ignore scripts_cache

  #2883 merged Nov 29, 2024

• Use tag pr-xyz for triton-benchmarks runs on PRs.

  #2880 merged Nov 29, 2024

• Merge OpenAI Triton commit cc89dac

  #2876 merged Nov 29, 2024

• Support for tt.dot_scaled operator

  #2804 merged Nov 29, 2024

• Generalize Intel coalescing pass to handle users of scf.for with coalesced load

  #2856 merged Nov 29, 2024

• Remove duplication of usage of XPU specific libraries in bin/CMakeLists.txt

  #2874 merged Nov 29, 2024

• [github-bot] Update spirv-llvm-translator.conf

  #2872 merged Nov 29, 2024

• [NFC] Remove CMAKE_VERBOSE_MAKEFILE var

  #2871 merged Nov 29, 2024

• [NFC] Remove get_event_pool from XPUUtils

  #2870 merged Nov 28, 2024

• Remove packaging from dependencies

  #2866 merged Nov 28, 2024

• [GEMM] Remove TRITON_INTEL_ENABLE_ADDRESS_PAYLOAD_OPT

  #2861 merged Nov 28, 2024

• Reapply "[BUILD] Some CMake cleanup/modernisation (#5271)"

  #2864 merged Nov 28, 2024

• Add small 2D load block size option for B.T matrix.

  #2863 merged Nov 28, 2024

• Change the order of bitcast op during the DPAS lowering to improve the instruction scheduling.

  #2860 merged Nov 28, 2024

• Improve performance of shape 1024x1024x1024 out of box

  #2839 merged Nov 28, 2024

• [github-bot] Update spirv-llvm-translator.conf

  #2857 merged Nov 28, 2024

• Merge OpenAI Triton commit 6d3ed0b

  #2858 merged Nov 28, 2024

• Revert some changes for Windows build since they seem unnecessary

  #2847 merged Nov 27, 2024

• Merge OpenAI Triton commit 9e508a4

  #2848 merged Nov 27, 2024

• Replace no-basekit.yml with pip-test.yml

  #2851 merged Nov 27, 2024

• Leave only one ruff-pre-commit hook in .pre-commit-config.yaml

  #2852 merged Nov 27, 2024

• Enable 08-grouped-gemm on A770

  #2849 merged Nov 27, 2024

• Replace "use_system_python" with "use_pyenv_python"

  #2842 merged Nov 27, 2024

• [XPU][TritonIntelGPUToLLVM] Add support for more transpose kinds

  #2786 merged Nov 27, 2024

• Make Windows build C++17 compliant

  #2833 merged Nov 27, 2024

• [XPU][OptRed] Revamp -tritonintelgpu-optimize-reduction-locality

  #2800 merged Nov 27, 2024

• Merge OpenAI Triton commit 3f1d70f

  #2838 merged Nov 26, 2024

• Fix test_chained_reductions

  #2821 merged Nov 26, 2024

• Generalize test_libdevice.py; don't import libdevice from extra.intel

  #2832 merged Nov 26, 2024

• Use [[maybe_unused]] to get rid of warnings while building SPIRVRunner

  #2829 merged Nov 26, 2024

• Fix test_scan_layouts[True-1-src_layout10-64-32] on LTS driver; ignoring loadBinary error with large registers

  #2808 merged Nov 26, 2024

• Return elapsed_time patch

  #2828 merged Nov 26, 2024

• [XPU][TritonIntelGPUToLLVM] Add support for more shuffle kinds

  #2799 merged Nov 26, 2024

• [NIT] Revert Allocation analysis changes

  #2826 merged Nov 26, 2024

• [XPU][TritonGPUToLLVM] Use llvm.func attributes to express kernels ND-ranges

  #2770 merged Nov 26, 2024

• Sync from upstream

  #2820 merged Nov 26, 2024

• Merge OpenAI Triton commit 22e212b

  #2819 merged Nov 25, 2024

• Update PyTorch commit id

  #2818 merged Nov 25, 2024

• Use find_package(Threads) instead of hardcoding

  #2813 merged Nov 25, 2024

• [XPU][TritonGEN] Replace split barrier ops usages with SPIR-V ops

  #2814 merged Nov 25, 2024

• Revert "bump triton verion to 3.1.0 (#2716)"

  #2812 merged Nov 25, 2024

• Workflow to test Triton with pip dependencies

  #2806 merged Nov 24, 2024

• Merge OpenAI Triton commit e3ab295

  #2810 merged Nov 24, 2024

• [github-bot] Update spirv-llvm-translator.conf

  #2809 merged Nov 23, 2024

• Merge OpenAI Triton commit 16ce143

  #2807 merged Nov 23, 2024

10 Pull requests opened by 7 people

• Fix headers path for conda run

  #2816 opened Nov 25, 2024

• Use order from A matrix when determining DPAS layout

  #2834 opened Nov 26, 2024

• [XPU][TritonGPUToLLVM] Use `reqd_work_group_size`

  #2845 opened Nov 27, 2024

• [XPU] Enable reduction optimization by default

  #2846 opened Nov 27, 2024

• [UT] Fix test_print UTs

  #2867 opened Nov 28, 2024

• Windows debug

  #2875 opened Nov 29, 2024

• [XPU] Drop `-tritonintelgpu-optimize-elementwise-locality` pass

  #2877 opened Nov 29, 2024

• Add `setuptools` to runtime deps

  #2881 opened Nov 29, 2024

• [CI][GEMM][FA] Remove default path with env vars configuration

  #2882 opened Nov 29, 2024

• [github-bot] Update spirv-llvm-translator.conf

  #2884 opened Nov 29, 2024

26 Issues closed by 8 people

• [CI] Specify `TAG` when triggering benchmarks runs on PRs

  #2862 closed Nov 29, 2024

• Merge OpenAI Triton till Nov 29th

  #2682 closed Nov 29, 2024

• Implement support for the `tt.dot_scaled` operation on XPU

  #2633 closed Nov 29, 2024

• Generalize Intel coalescing pass to handle users of scf.for with coalesced load

  #2762 closed Nov 29, 2024

• [GEMM] 2048x2048x2048 out of box performance degradation

  #2733 closed Nov 29, 2024

• [GEMM] 16384x8192x4096 out of box performance degradation

  #2734 closed Nov 29, 2024

• Consider remove `packaging` from dependencies

  #2865 closed Nov 28, 2024

• Reland upstream commit `2003685`

  #2859 closed Nov 28, 2024

• [GEMM] Improve performance of shape 1024x1024x1024 out of box

  #2822 closed Nov 28, 2024

• Try to enable `bandit` pre-commit check in Triton

  #2844 closed Nov 27, 2024

• Leave only one `ruff-pre-commit` hook in `.pre-commit-config.yaml`

  #2854 closed Nov 27, 2024

• Tutorial 08-grouped-gemm hangs on A770

  #2737 closed Nov 27, 2024

• Allow sub-group transpose and shuffles with more than one contiguous row per thread

  #2749 closed Nov 27, 2024

• Make Windows build C++17 compliant

  #2837 closed Nov 27, 2024

• Try ro port `test_inductor_cummax_bool` to Triton

  #2841 closed Nov 27, 2024

• Revamp `-tritonintelgpu-optimize-reduction-locality`

  #2752 closed Nov 27, 2024

• Try to port `test_bessel` to Triton

  #2835 closed Nov 27, 2024

• Generalize code in `test_gpuhello.py`

  #2840 closed Nov 27, 2024

• AttributeError: module 'triton' has no attribute 'jit'

  #2740 closed Nov 27, 2024

• [UT] Fix `test_chained_reductions[in_shape0-perm0-red_dims0]`

  #2703 closed Nov 26, 2024

• Fix `test_scan_layouts[True-1-src_layout10-64-32]` on LTS driver

  #2789 closed Nov 26, 2024

• Upstream changes for allocation

  #2825 closed Nov 26, 2024

• [GEMM] Improve performance of shape 8192x16384x4096 out of box

  #2796 closed Nov 26, 2024

• Investigate the new tensor descriptor API

  #2586 closed Nov 25, 2024

• Replace SplitBarrierSignalOp TritonGEN operations with SPIR-V dialect ones

  #2815 closed Nov 25, 2024

• Workflow to test Triton with runtime dependencies installed with pip

  #2795 closed Nov 24, 2024

13 Issues opened by 7 people

• Merge OpenAI Triton till Dec 13rd

  #2879 opened Nov 29, 2024

• [UT] Fix `test_gather`

  #2878 opened Nov 29, 2024

• [GEMM] 2048X2048X2048 has big variance.

  #2873 opened Nov 29, 2024

• Attach `opencl.kernels` metadata

  #2869 opened Nov 28, 2024

• [Pytorch Upstream] Triton build failed with docker image pytorch/manylinux2_28-builder:cpu

  #2868 opened Nov 28, 2024

• `'/debug:fastlink'` and `'/INCREMENTAL'` aren't recognized in Windows build

  #2855 opened Nov 27, 2024

• Drop `-tritonintelgpu-optimize-elementwise-locality` pass

  #2850 opened Nov 27, 2024

• SPIRVRunner: Investigate and handle multi-kernel execution (benchmark- gemm_streamk_benchmark.py)

  #2831 opened Nov 26, 2024

• Failed core tests on A50

  #2830 opened Nov 26, 2024

• Reduce changes in common files for windows support

  #2824 opened Nov 26, 2024

• Remove `getElemsPerThreadForOperands` from `MmaEncodingTrait`

  #2823 opened Nov 26, 2024

• Flaky Triton build

  #2817 opened Nov 25, 2024

• Reland upstream commit `340cbc6`

  #2811 opened Nov 23, 2024

10 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• [DOCUMENTS]Update the DPAS encoding documents.

  #2746 commented on Nov 28, 2024 • 1 new comment

• Insert freeze between masked loads and sdiv/srem instructions

  #2775 commented on Nov 28, 2024 • 1 new comment

• [UT] regression in test_subprocess.py with the PTDB 0.5.3

  #800 commented on Nov 26, 2024 • 0 new comments

• Classify difference between Intel port and OpenAI Triton

  #2030 commented on Nov 26, 2024 • 0 new comments

• Enable all tests from `test_debug.py` on XPU

  #2755 commented on Nov 28, 2024 • 0 new comments

• Cleanup unnecessary env variables

  #1972 commented on Nov 28, 2024 • 0 new comments

• Implement support for `TritonGPU::UpcastMXFPOp` for Intel XPU BE

  #2678 commented on Nov 29, 2024 • 0 new comments

• [GEMM] Improve shapes with performance <95% of XeTLA

  #2024 commented on Nov 29, 2024 • 0 new comments

• POC: Enable Proton for XPU

  #2635 commented on Nov 28, 2024 • 0 new comments

• Support for `tritongpu.upcast_mxfp` operation

  #2700 commented on Nov 26, 2024 • 0 new comments