Halide Files

Release highlights

We have deliberately skipped version 20.0.0 to align with the LLVM version we are now using. Note that LLVM 21.1.1 or higher is required as LLVM 21.1.0 has a major bug in the NVPTX backend.

Major changes

• The rfactor scheduling directive was rewritten and enhanced. It is now compatible with autoschedulers.
• The Mullapudi2016 autoscheduler now supports experimental GPU scheduling.
• The Python bindings have been substantially improved, with many missing bindings filled in.
• HL_DEBUG_CODEGEN gained a new filtering mode. Debug levels can now be set on a per-file/per-function basis.
• Support was added for AMD Zen5 and the iOS Simulator.
• The strict_float feature has been reimplemented and should be much more reliable.
• Lots of bugfixes, performance improvements, and build system improvements. We spent a lot of time fixing issues with our testing infrastructure and are looking forward to implementing a more stable contribution experience going forward.

Deprecations

• LLVM 19 and below are no longer supported, in keeping with our support policy.
• Halide_BUNDLE_STATIC will be removed in the next release. If you are using it, please migrate to the shared library instead.
• Support for Python 3.8 has been dropped.

Changelog

Scheduling

• The rfactor scheduling directive was rewritten and enhanced.
• Rewrite the rfactor scheduling directive by @alexreinking in https://github.com/halide/Halide/pull/8490
• Dequalify names when constructing RVars in rfactor by @alexreinking in https://github.com/halide/Halide/pull/8560
• Add promise_clamped in rfactor by @alexreinking in https://github.com/halide/Halide/pull/8608
• Add rfactor patterns for NaN-propagating min/max by @alexreinking in https://github.com/halide/Halide/pull/8587
• The Mullapudi2016 autoscheduler now supports experimental GPU scheduling.
• GPU autoscheduling with Mullapudi2016: the reference implementation by @antonysigma in https://github.com/halide/Halide/pull/7787
• Mullapudi2016-GPU: Reorder to avoid for-loops to be sandwiched between gpu_blocks. by @antonysigma in https://github.com/halide/Halide/pull/8647
• Enable experimental Mullapudi2016 GPU scheduler for test-bench by @antonysigma in https://github.com/halide/Halide/pull/8650
• Highlight Metal GPU code in stmt_html by @antonysigma in https://github.com/halide/Halide/pull/8659
• Always ensure gpu_threads count >= warp size of 32 by @antonysigma in https://github.com/halide/Halide/pull/8656
• Fix incorrect natural vector size on Zen4 by @abadams in https://github.com/halide/Halide/pull/8570
• Make it an error to use a device extern stage without target support by @abadams in https://github.com/halide/Halide/pull/8794
• Add support for adding tuple outputs in the configure() method by @abadams in https://github.com/halide/Halide/pull/8649

Python

• Fix argument order in rpow by @alexreinking in https://github.com/halide/Halide/pull/8677
• Drop support for Python 3.8 by @alexreinking in https://github.com/halide/Halide/pull/8678
• Fix segfault in RDom's operator<< by @alexreinking in https://github.com/halide/Halide/pull/8679
• Use ruff to format and lint Python code by @alexreinking in https://github.com/halide/Halide/pull/8684
• Get raw Runtime::Buffer from Buffer in Python rather than use PyBuffer by @alexreinking in https://github.com/halide/Halide/pull/8682
• Bind in-place update operators (e.g. +=) in Python by @alexreinking in https://github.com/halide/Halide/pull/8683
• Clean up Python dependencies; document uv usage by @alexreinking in https://github.com/halide/Halide/pull/8694
• Fix several printing segfaults. by @alexreinking in https://github.com/halide/Halide/pull/8700
• Add Python bindings for serialization by @alexreinking in https://github.com/halide/Halide/pull/8718
• Add all remaining IROperator ops to Python bindings by @alexreinking in https://github.com/halide/Halide/pull/8771
• Fix up memoize; bind to Python by @alexreinking in https://github.com/halide/Halide/pull/8778
• Fix invalid Python type annotation and return types (#8772) by @rtzam in https://github.com/halide/Halide/pull/8773
• Expose Runtime::Buffer::cropped to C++ and Python Buffer by @rtzam in https://github.com/halide/Halide/pull/8787

Debugging

• New feature flag to allow for stack backtrace/unwind by @mcourteaux in https://github.com/halide/Halide/pull/8703
• Add filtering capabilities to HL_DEBUG_CODEGEN by @alexreinking in https://github.com/halide/Halide/pull/8627
• Adding worker_thread_idle() for more informative profiling by @slomp in https://github.com/halide/Halide/pull/8719
• Color IR output in cout and cerr. by @mcourteaux in https://github.com/halide/Halide/pull/8635
• Improve output format for lowering passes timing. by @mcourteaux in https://github.com/halide/Halide/pull/8749
• fix(stmt-html): Fix embedded Buffer processing performance issue. by @mcourteaux in https://github.com/halide/Halide/pull/8748
• Use AArch64 assembly syntax on macOS with LLVM<22 by @alexreinking in https://github.com/halide/Halide/pull/8710

CodeGen

• Mark our PTX kernels as kernels, to stop them from being stripped by @abadams in https://github.com/halide/Halide/pull/8571
• Math functions renaming table for GPU backends to support vectorized evaluation of math functions. by @mcourteaux in https://github.com/halide/Halide/pull/8595
• Apply version constraints to iOS objects by @alexreinking in https://github.com/halide/Halide/pull/8546
• Redirect bitwise ops to logical ops in case the arguments are bool. by @mcourteaux in https://github.com/halide/Halide/pull/8597
• scalarize select condition for LLVM where possible by @abadams in https://github.com/halide/Halide/pull/8575
• Add missing addition simplifier rules by @abadams in https://github.com/halide/Halide/pull/8630
• Bounds and alignment analysis through bitwise ops by @abadams in https://github.com/halide/Halide/pull/8574
• Make the vld2 pattern more obviously profitable by @abadams in https://github.com/halide/Halide/pull/8765
• Fix vector shuffle for Vulkan CodeGen by @derek-gerstmann in https://github.com/halide/Halide/pull/8621
• Suppress warning on Windows for duplicate constant symbols. by @mcourteaux in https://github.com/halide/Halide/pull/8555
• Use lossless_cast for saturating casts from unsigned to signed on x86 by @abadams in https://github.com/halide/Halide/pull/8527
• AMD Zen5 support by @changhoon-sung in https://github.com/halide/Halide/pull/8612

Compiler

• Rework strict_float to use individual op intrinsics instead by @abadams in https://github.com/halide/Halide/pull/8641
• Don't cache mutations of Exprs that have only one reference to them by @abadams in https://github.com/halide/Halide/pull/8518
• Only use the nodes-visited set for nodes with multiple refs by @abadams in https://github.com/halide/Halide/pull/8547
• In graph_equal(), call the correct implementation for comparing equalities between statements and expressions by @BachiLi in https://github.com/halide/Halide/pull/8611

Runtime

• Support copying the overlapping region from one buffer to another. by @mcourteaux in https://github.com/halide/Halide/pull/8463
• Add (iOS) simulator target feature. by @alexreinking in https://github.com/halide/Halide/pull/8623
• Opt out of JIT exceptions by @abadams in https://github.com/halide/Halide/pull/8615
• Experimental: support removing unused runtime functions via HL_RUNTIME_DROP_FUNCS environment variable.
• PoC feature: drop functions from the runtime by @mcourteaux in https://github.com/halide/Halide/pull/8653

Apps

• The onnx app now builds with CMake:
• Add CMake for onnx app by @vawale in https://github.com/halide/Halide/pull/8707
• Fix halide_as_onnx_backend_test by @alexreinking in https://github.com/halide/Halide/pull/8784

Documentation

• Add note about relative paths to readme by @abadams in https://github.com/halide/Halide/pull/8613

Bugfixes

• Fix [#8534] [Buffer serialization does not match deserialization] by @abadams in https://github.com/halide/Halide/pull/8535
• Fix CUDA HTML code printing bug. by @mcourteaux in https://github.com/halide/Halide/pull/8558
• Fix halide_get_cpu_features() linkage to avoid name mangling issues by @derek-gerstmann in https://github.com/halide/Halide/pull/8573
• Fix for [#8578] by @mcourteaux in https://github.com/halide/Halide/pull/8579
• Fix shuffle bug in CodeGen C. by @mcourteaux in https://github.com/halide/Halide/pull/8567
• Check if expression is defined before trying to compute its constant_integer_bounds by @vksnk in https://github.com/halide/Halide/pull/8599
• Drop invalid "in-bounds" GEP for constant offsets by @alexreinking in https://github.com/halide/Halide/pull/8768
• Record trace_loads directly on ImageParam. by @alexreinking in https://github.com/halide/Halide/pull/8803
• RewriteLoadsAs32Bit should use the mutated index by @rootjalex in https://github.com/halide/Halide/pull/8581
• Set any_strict_float for wrapper module if target has strict_flag feature by @vksnk in https://github.com/halide/Halide/pull/8681
• Fix wrong type of the bound by @vksnk in https://github.com/halide/Halide/pull/8781
• Fix UB-introducing rewrite in FindIntrinsics by @abadams in https://github.com/halide/Halide/pull/8539
• Fix rewrite that doesn't preserve type by @abadams in https://github.com/halide/Halide/pull/8674
• Fix nested select handling in remove_undef by @abadams in https://github.com/halide/Halide/pull/8669
• Add an underlying type to the halide_buffer_flags to prevent UB in C++ by @mcourteaux in https://github.com/halide/Halide/pull/8690

Testing / CI

• Limit depth more strictly in CSE fuzz test by @abadams in https://github.com/halide/Halide/pull/8512
• Skip fast exp/log/pow/sin/cosine tests without sse 4.1 by @abadams in https://github.com/halide/Halide/pull/8541
• Hopefully fix flaky mullapudi reorder test by @abadams in https://github.com/halide/Halide/pull/8542
• Skip test when code could be using x87 by @abadams in https://github.com/halide/Halide/pull/8537
• Fix stale GPU lifetime management tests for Vulkan. by @derek-gerstmann in https://github.com/halide/Halide/pull/8601
• Upgrade runner for cmake_cmake_file_lists job by @alexreinking in https://github.com/halide/Halide/pull/8609
• Buildbot fixes by @alexreinking in https://github.com/halide/Halide/pull/8706
• Fix the pip packaging workflow by @alexreinking in https://github.com/halide/Halide/pull/8708
• Fix complexity of bounds of nested pure intrinsics by @abadams in https://github.com/halide/Halide/pull/8689
• Skip two sub-tests on llvm 21.1 by @abadams in https://github.com/halide/Halide/pull/8782
• Speed up simd_op_check_wasm by @abadams in https://github.com/halide/Halide/pull/8780
• Reduce the beam size in the adams2019 apps test to avoid timeouts by @abadams in https://github.com/halide/Halide/pull/8786
• Workaround llvm slow compile time bug in Mullapudi overlap test by @abadams in https://github.com/halide/Halide/pull/8793
• Restore concurrent behavior to gpu_allocation_cache test by @abadams in https://github.com/halide/Halide/pull/8792
• Revert "Skip two sub-tests on llvm 21.1" by @abadams in https://github.com/halide/Halide/pull/8806
• Fix WASM splat op check test. by @mcourteaux in https://github.com/halide/Halide/pull/8705

Build

• Fix workflow for next release by @alexreinking in https://github.com/halide/Halide/pull/8514
• Fix Debian packaging by @alexreinking in https://github.com/halide/Halide/pull/8524
• Remove llvm version check from Makefile by @abadams in https://github.com/halide/Halide/pull/8533
• Drop deprecated / unsupported setups for Halide 20 by @alexreinking in https://github.com/halide/Halide/pull/8508
• Fix check for Windows never having aligned_alloc available. by @mcourteaux in https://github.com/halide/Halide/pull/8551
• Don't include CMAKE_INSTALL_PREFIX when LIBDIR is absolute by @alexreinking in https://github.com/halide/Halide/pull/8552
• Add target-nvptx to target-all in vcpkg.json by @alexreinking in https://github.com/halide/Halide/pull/8562
• Fix top of LLVM, and remove upper limit of LLVM version from CMakeLists. by @mcourteaux in https://github.com/halide/Halide/pull/8568
• build_halide_h asserts that every header it slurps in is one of the args by @abadams in https://github.com/halide/Halide/pull/8559
• Upgrade pybind11 to 2.11.1 by @alexreinking in https://github.com/halide/Halide/pull/8616
• Drop check for LLVM_LIBCXX in FindHalide_LLVM.cmake by @alexreinking in https://github.com/halide/Halide/pull/8617
• Fix finding LLD on Homebrew when multiple versions are installed. by @alexreinking in https://github.com/halide/Halide/pull/8619
• Fix build on GCC 15 (Comes with Fedora 42). by @mcourteaux in https://github.com/halide/Halide/pull/8626
• Constrain Clang and LLD searches to LLVM version by @alexreinking in https://github.com/halide/Halide/pull/8634
• Disallow empty CMAKE_BUILD_TYPE on single-config generators by @alexreinking in https://github.com/halide/Halide/pull/8651
• Add missing outputs to add_halide_library; fix advice in Lesson 21. by @alexreinking in https://github.com/halide/Halide/pull/8660
• Bump the LLVM version in the pip package to 20.1.8 by @alexreinking in https://github.com/halide/Halide/pull/8698
• Prefer to build against libjpeg-turbo and document this. by @alexreinking in https://github.com/halide/Halide/pull/8775
• Add C++17 requirement to RunGenMain CMake target by @alexreinking in https://github.com/halide/Halide/pull/8795
• Allow llvm-ar in BundleStatic.cmake by @alexreinking in https://github.com/halide/Halide/pull/8799
• Fix dubious find_package logic in test/generator by @alexreinking in https://github.com/halide/Halide/pull/8804
• Warning when extra-output is requested w/o filename by @FabianSchuetze in https://github.com/halide/Halide/pull/8671
• Makefile linker flag fixes and cleanups by @abadams in https://github.com/halide/Halide/pull/8764

Ongoing maintenance

• Fix clang-tidy-19 errors by @steven-johnson in https://github.com/halide/Halide/pull/8509
• Remove unused function in HexagonOptimize by @steven-johnson in https://github.com/halide/Halide/pull/8511
• Fix two non-idiomatic uses of node_type by @abadams in https://github.com/halide/Halide/pull/8520
• Handle some misc TODOs by @abadams in https://github.com/halide/Halide/pull/8528
• Use a consistent idiom for visit_let by @abadams in https://github.com/halide/Halide/pull/8540
• Upgrade to clang-format 19 by @alexreinking in https://github.com/halide/Halide/pull/8543
• Suppress clang-tidy warning for make_with_shape_of() by @steven-johnson in https://github.com/halide/Halide/pull/8545
• Remove debugging print left in by @abadams in https://github.com/halide/Halide/pull/8572
• Fixes for llvm trunk by @abadams in https://github.com/halide/Halide/pull/8590
• Another fix for llvm trunk by @abadams in https://github.com/halide/Halide/pull/8591
• Our internal error macros were redesigned:
• Accurately annotate Error system with [[noreturn]] by @alexreinking in https://github.com/halide/Halide/pull/8564
• Teach compilers that internal_error does not return. by @alexreinking in https://github.com/halide/Halide/pull/8807
• Use a new macro trick to avoid throwing in destructors. by @alexreinking in https://github.com/halide/Halide/pull/8774
• Move to opaque llvm pointers by @abadams in https://github.com/halide/Halide/pull/8614
• Avoid throwing from a destructor in PartitionLoops.cpp by @alexreinking in https://github.com/halide/Halide/pull/8767
• Bump version to 21.0.0 by @alexreinking in https://github.com/halide/Halide/pull/8810
• remove superfluous overload that causes compile errors by @ongjunjie in https://github.com/halide/Halide/pull/8654
• Remove obsolete WasmExecutor specific debug macro. by @zvookin in https://github.com/halide/Halide/pull/8670
• Add missing header by @vksnk in https://github.com/halide/Halide/pull/8680
• Attempted fix for LLVM change by @abadams in https://github.com/halide/Halide/pull/8642
• Fix top LLVM: renamed NVPTX barrier intrinsics. by @mcourteaux in https://github.com/halide/Halide/pull/8631

New Contributors

• @changhoon-sung made their first contribution in https://github.com/halide/Halide/pull/8612
• @vawale made their first contribution in https://github.com/halide/Halide/pull/8707
• @rtzam made their first contribution in https://github.com/halide/Halide/pull/8773

Full Changelog: https://github.com/halide/Halide/compare/v19.0.0...v21.0.0