taichi

Full changelog:

• [ci] Fix docker prebuilt binary link (#2483) (by Ye Kuang)
• [wasm] Implement materialize() for wasm backend and clean up unused functions (#2480) (by squarefk)
• [misc] Unify std::filesystem header (#2478) (by Ye Kuang)
• [opengl] Dump compute shader source code when print_kernel_llvm_ir is true (#2479) (by xndcn)
• [metal] Fix randseedoffset_in_runtime_buffer (#2477) (by Ye Kuang)
• [Metal] Support pointer SNode on Metal (#2441) (by Ye Kuang)
• [wasm] Recover code (#2476) (by ljcc0930)
• [AutoDiff] Automatically determine AdStack's size (#2438) (by xumingkuan)
• [misc] Add checking of gitpython in cmake (#2473) (by xndcn)
• [ci] Restrict slash-command-dispatch to PR only (#2472) (by Ye Kuang)
• [ir] Use SNodeTree to implement root (#2449) (by Ye Kuang)
• [Lang] Fix parameter conflict between ti.maybe_transform_ti_func_call_to_stmt() and ti.external_func_call() (#2470) (by squarefk)

Full changelog:

• [ci] Fix docker caching issue (#2468) (by Ye Kuang)
• [ci] Separate Win build and test steps (#2465) (by Ye Kuang)
• [misc] Fix parameters' type. (#2464) (by ljcc0930)
• [wasm] [aot] Support AOT on webassembly backend. (#2461) (by ljcc0930)

Full changelog:

• [docker] Add twine installation in dockerfile (#2458) (by Jiasheng Zhang)
• [WASM] Add set_root() for each WASM module (#2429) (by squarefk)
• [ci] [docker] Build release with docker in ubuntu18.04 (#2457) (by Jiasheng Zhang)
• [Metal] [Aot] Fix dumped runtime buffer size (#2454) (by Rachel Gu)
• [Doc] Document lazy_grad() function (#2456) (by FantasyVR)
• [misc] Update prtags.json to include aot (#2455) (by Rachel Gu)
• [bug] Fix linux detection macro (#2453) (by Ye Kuang)
• [bug] Fix assertion failure when there is a redundant cast (#2451) (by xumingkuan)
• Set process DPI awareness when detecting system version less than 6.3 (Windows 8.1) (#2450) (by Bob Cao)
• [ir] [bug] Make control-flow graph take function call into account (#2448) (by xumingkuan)

Full changelog:

• [ir] [bug] Skip test_template (#2444) (by Ye Kuang)
• [metal] Always initialize root SNode ListManager (#2443) (by Ye Kuang)
• [metal] Separate kernel utils and AOT utils (#2440) (by Ye Kuang)
• [ir] Add more comments to CFG optimizations and analyses (#2437) (by xumingkuan)
• [Metal] Make Aot module dumps buffer size (#2431) (by Rachel Gu)
• [ir] Support BLS in control-flow graph and add docstrings in build_cfg (#2412) (by xumingkuan)
• [Lang] [refactor] Separate runtime and root initialization (#2424) (by Ye Kuang)
• [ci] Moved windows build & test to github action (#2436) (by Jiasheng Zhang)
• [metal] Add 3-stage GC Metal kernels (#2268) (by Ye Kuang)
• [CUDA] [bug] Fix unified memory allocator when running on non-main thread (#2427) (by Ye Kuang)
• [CI] Fixed windows build & test (#2434) (by Jiasheng Zhang)
• [ci] Trigger Build and Test (CPU) after the required checks (#2435) (by Ye Kuang)
• [Bug] Fix 'ti.get_addr' failures in async mode (#2430) (by Dunfan Lu)
• [Lang] Add get_addr() to retrieve the buffer addr of an SNode (#2426) (by Dunfan Lu)
• [IR] Fully deprecate trailing_bits (#2422) (by Kenneth Lozes)
• [ir] Move infer_snode_properties out from StructCompiler (#2420) (by Ye Kuang)
• [doc] Add documentation for IndexExtractor (#2416) (by Ye Kuang)
• [Doc] Changed llvm-10 link and installation instruction (#2419) (by Jiasheng Zhang)
• [wasm] [llvm] Support wasm with single thread (#2415) (by squarefk)
• [ir] Add CHI Builder code examples (#2383) (by bx2k)
• [ci] Remove travis configs (#2414) (by Ye Kuang)
• [opt] [bug] Simplify modulo POT (Stage 2) (#2411) (by xumingkuan)
• [Bug] Fixed internal lagrangian forces (#2413) (by Saullo G P Castro)
• [lang] Add override to fix compilation warnings (#2410) (by Ye Kuang)
• [ir] Make lower_scalar_pointer testable (#2407) (by Ye Kuang)

Full changelog:

• [CI] Modified build.py to use token for all upload (#2408) (by Jiasheng Zhang)
• [CI] Configured Jenkinsfile to support nightly test (#2405) (by Jiasheng Zhang)
• [ir] Add random and atomic statements to IR Builder (#2379) (by bx2k)
• [lang] Fix filesystem header not found on older GCC (#2401) (by Ye Kuang)
• [Metal] Make file names dumped from aot configurable (#2402) (by Rachel Gu)
• [Misc] Fixed that mass_spring_3d.py do not exit on import error (#2400) (by Jiasheng Zhang)
• [lang] Add tests for refine_coordinates (#2382) (by Ye Kuang)
• [CI] Simplified github CI procudure. (#2399) (by Jiasheng Zhang)
• [CI] Support auto cancel workflows of previous commits (#2397) (by Jiasheng Zhang)
• [CI] Recovered format check, add dependancy (#2394) (by Jiasheng Zhang)
• [Opt] Optimize nested type casts (#2390) (by xumingkuan)
• [type] [bug] Remove redundant component of bit pointer struct (#2393) (by Jiafeng Liu)
• [CI] Add comment dispatch in pull request to support auto format (#2392) (by Jiasheng Zhang)
• [Doc] Add more documentation regarding fast_gui (#2385) (by Ye Kuang)
• [metal] Fix Nonetype due to un-materialized runtime (#2389) (by Rachel Gu)
• [Doc] Update developer installation doc regarding LLVM_DIR (#2384) (by Ye Kuang)
• [metal] Add AOT module builder for metal (#2372) (by Ye Kuang)
• [ir] Add nodiscard to get_loop_guard/get_if_guard (#2377) (by xumingkuan)
• [vulkan] Add Vulkan API (#2376) (by Ye Kuang)
• [Lang] [ir] Consecutive indices for all snodes (#2327) (by Kenneth Lozes)
• [Misc] Remove unnecessary scipy dependency in misc folder (#2373) (by Daniel)
• [lang] Add AOT module interface (#2364) (by Ye Kuang)
• [metal] Fix metal f32 bug in the print buffer (#2368) (by Rachel Gu)
• [Doc] Change installer for archlinux from yaourt to yay (#2261) (by liberix)
• [CI] Added issue_comment.yml to handle PR reply triggered actions (#2370) (by Jiasheng Zhang)
• [misc] Add CI tag (#2371) (by Ye Kuang)
• [Opt] Add error if bls analysis failed (#2366) (by Jack12xl)
• [Misc] Modified build.py to better execute release (#2365) (by Jiasheng Zhang)
• [lang] Fix serialization key value pairs (#2363) (by Ye Kuang)
• [Misc] Add test flag when releasing taichi (#2362) (by Jiasheng Zhang)
• [Example] Improve stable fluids (#2361) (by Rachel Gu)
• [lang] Remove serializer's variadic template API (#2358) (by Ye Kuang)

Full changelog:

• [ir] Remove "kernel" when printing ReturnStmt (#2353) (by xumingkuan)
• [misc] Support Python 3.9 (#2274) (by Derived Cat)
• [opt] Simplify modulo POT (Stage 1) (#2352) (by xumingkuan)
• [opt] Slightly simplify algebraic simplification (#2337) (by xumingkuan)
• [Refactor] Add a class Callable to unify Kernel and Function (#2338) (by xumingkuan)
• [ir] Rename KernelReturnStmt to ReturnStmt (#2349) (by xumingkuan)
• [lang] Support std::optional and enum classes in the serializer (#2350) (by Ye Kuang)
• [IR] Support DiffRange shl (#2346) (by Kenneth Lozes)
• [Docker] Fix the docker build by bumping CMake version in Dockerfile (#2348) (by Chengchen(Rex) Wang)
• [ir] [test] Add tests for make_block_local (#2343) (by Ye Kuang)
• [ir] Add ArithmeticInterpretor to evaluate a subset of CHI IR (#2342) (by Ye Kuang)
• [opt] Better encapsulate BLS bounds (#2341) (by Ye Kuang)
• [Example] cornell_box: Remove unused sphere (#2334) (by Aaryaman Vasishta)
• [opt] Simplify replace_statements and improve demote_dense_struct_fors (#2335) (by xumingkuan)
• [refactor] [test] Reorganize tests' file structure (#2336) (by xumingkuan)

Full changelog:

• [IR] Experimental real function support (Stage 1) (#2306) (by xumingkuan)
• [Opt] Simplify multiplying/dividing POT (#2332) (by xumingkuan)
• [Lang] Add ti.sym_eig for 2x2 matrices (#2328) (by Robslhc)
• [perf] Loop-invariant code motion (#2323) (by Bob Cao)
• [bug] fix potential bug in test_eig.py (#2329) (by Robslhc)
• [example] Colored triangle rasterizer (#2315) (by Bob Cao)
• [doc] Windows uses backslash for filepath (#2319) (by Bob Cao)
• [gui] Set DPI awareness context to per-monitor-aware to create pixel perfect windows without system scaling. (#2320) (by Bob Cao)
• [doc] wrap for to avoid confusion (#2322) (by Ravenclaw-OIer)
• [ir] Make BLSAnalyzer testable (#2294) (by Ye Kuang)
• [lang] Add rescale_index() and test (#2313) (by Kenneth Lozes)
• [bugfix] Protect against path with space. (#2318) (by Bob Cao)
• [LLVM] Add random seed support (#2297) (by Andrew Sun)
• [Lang] Add ti.eig for 2x2 matrices (#2303) (by Robslhc)
• support diffRange multiplication (#2310) (by Kenneth Lozes)

Full changelog:

• [misc] Set up buildbot for gpu tests on linux (#2305) (by bx2k)
• [doc] Update sparse computation doc (#2302) (by Yuanming Hu)

Full changelog:

• [Lang] Add ti.randn (#2266) (by Andrew Sun)
• [ir] [refactor] Rename StackXStmt to AdStackXStmt (#2283) (by xumingkuan)
• [ir] Rename set_arg_nparray to set_arg_external_array (#2280) (by xumingkuan)
• [ir] [refactor] Remove OffloadedStmt::step (#2282) (by xumingkuan)
• [test] Add a test for range analysis of indices of reversed loops (#2279) (by xumingkuan)
• [refactor] Remove legacy C++ frontend macros (#2278) (by Robslhc)
• [IR] Add inline docs to all statements (#2276) (by xumingkuan)
• [ir] Rename is_np_array to is_external_array except the frontend (#2277) (by xumingkuan)
• [opt] Avoid recursively generating indices twice for BLS (#2272) (by xumingkuan)
• [refactor] Unified ti_core usage to _ti_core in python/taichi/misc, python/taichi/main and testing. (#2270) (by Jiasheng Zhang)
• [misc] Update README.md (#2269) (by Yuanming Hu)
• [ir] Move uniquely_accessed_bit_structs from compile_to_offloads to AnalysisManager (#2264) (by xumingkuan)
• [ir] Remove type_check in clone (#2262) (by xumingkuan)
• [type] [opt] Use BitStructStoreStmt for CustomFloatType with non-shared exponents (#2259) (by xumingkuan)

Full changelog:

• [IR] [refactor] Remove all get_kernel() and get_config() usages in IR passes (#2252) (by xumingkuan)
• [refactor] Rename Tape class to TapeImpl (#2258) (by Ye Kuang)
• [llvm] Well-behaved float64 random number generator for CPU and CUDA backends (#2253) (by Andrew Sun)
• [test] Rename algorithm simplification test to algebraic simplification test (#2256) (by xumingkuan)
• [Test] Migrate all cpp tests to googletest (#2238) (by bx2k)
• [IR] [refactor] Unify field names in load/store/atomic statements (#2250) (by xumingkuan)
• [test] [ir] Add tests for the alias_analysis pass (#2246) (by Ye Kuang)
• [misc] Fix typo in cmake error messages (#2249) (by xumingkuan)
• [misc] Sort python import order (#2247) (by Taichi Gardener)
• [lang] Replaced usage of taichi.lang.core with taichi.core.util.ti_core. (#2245) (by Jiasheng Zhang)
• [test] Add tests for value_diff_ptr_index pass (#2244) (by Ye Kuang)
• [ir] Add RAII guards to IR Builder (#2242) (by xumingkuan)
• [refactor] [ir] Rename "parallelize" to "num_cpu_threads" (#2243) (by xumingkuan)
• [ir] Add global variables to IR Builder (#2240) (by xumingkuan)
• [doc] Add installation instructions for Fedora and RHEL (#2239) (by Zhangyuan Nie)
• [refactor] Add ASTResolver to properly resolve AST nodes to functions (#2234) (by Ye Kuang)
• [Doc] Update doc typo (#2233) (by al)
• [refactor] Cleanup python imports (#2232) (by Ye Kuang)
• [refactor] Cleanup python imports (#2231) (by Ye Kuang)
• [test] [ir] [bug] Fix googletest on Windows and fix a bug in IR Builder (#2230) (by xumingkuan)
• [ir] Add more statements and tests to IR Builder (#2225) (by xumingkuan)
• [misc] Make taichi-gardener also run the required CI jobs (#2229) (by Ye Kuang)
• [refactor] Cleanup python imports (#2226) (by Ye Kuang)
• [example] Update mgpcg.py (#2228) (by qbao0x3273)
• [refactor] Clean up python import statements (#2224) (by Ye Kuang)
• [test] Enable C++ unit tests on CI (#2222) (by Ye Kuang)
• [test] Migrate test_alg_simp to googletest (#2220) (by Ye Kuang)
• [refactor] Create a library taichi_isolated_core that is free from pybind11 (#2218) (by Ye Kuang)
• [test] [ir] Add a simple unit test for IRBuilder (#2214) (by Ye Kuang)
• [ir] Add a constructor for the Kernel class to handle CHI IR (#2213) (by xumingkuan)

Full changelog:

• [refactor] Move TypedConstants to taichi/ir/type (#2211) (by Ye Kuang)
• [refactor] Move ASTBuilder and FrontendContext to frontend_ir (#2210) (by bx2k)
• [ir] [transforms] Added assertion that indices won't cause overflow under debug mode (#2199) (by Jiasheng Zhang)
• [refactor] Move code away from lang_utils (#2209) (by Ye Kuang)
• [refactor] Move type related utils away from lang_util.h (#2206) (by Ye Kuang)
• [refactor] Add SNode::GradInfoProvider to isolate SNode from Expr (#2207) (by Ye Kuang)
• [refactor] Separate SNode read/write kernels into a dedicated class (#2205) (by Ye Kuang)
• [IR] Add an IR Builder with some basic functions (#2204) (by xumingkuan)
• [test] Add a basic unit test using googletest (#2201) (by Ye Kuang)
• [refactor] Make taichi/common self contained (#2200) (by Ye Kuang)
• [ir] Generate yaml documentation for statement classes (#2192) (by xumingkuan)
• [test] Add googletest as a submodule (#2197) (by Ye Kuang)
• [test] [opengl] Avoid floor division cornor cases by adjusting test data (#2191) (by bx2k)
• [Bug] [lang] Fix AST not being transformed inside ti.ndrange (#2187) (by 彭于斌)

Full changelog:

• [misc] Refactor release workflow to upload the wheel artifacts (#2186) (by Ye Kuang)
• [test] Fix GPU testing on zhen (#2184) (by Yuanming Hu)
• [misc] Fix PyPi upload step with envs (#2182) (by Ye Kuang)

Changelog:

• [bug] Disable test_dense_dynamic for CUDA (#2179) (by Ye Kuang)
• [type] Apply quant_opt_atomic_demotion when storing all components (#2176) (by xumingkuan)
• [type] Improve bit struct store fusion and atomic demotion (#2175) (by xumingkuan)
• [type] Atomic demotion for bit struct stores (#2174) (by xumingkuan)
• [refactor] Move LLVM CustomType-related functions to codegen_llvm_quant.cpp (#2173) (by Taichi Gardener)
• [type] Use a single atomicCAS for BitStructStoreStmt (#2171) (by xumingkuan)
• [async] Add option async_opt_fusion_max_iter (#2170) (by Yuanming Hu)
• [async] Set the default value of async_flush_every to 50 (#2169) (by xumingkuan)
• [async] Add config flag async_max_fuse_per_task (#2165) (by Ye Kuang)
• [type] Support reading bit_struct as its physical type (#2166) (by Yuanming Hu)
• [infra] Support Timelines as a multithreading profiler (#2164) (by Yuanming Hu)
• [async] [lang] [opt] Add ti.loop_unique(covers=...) to improve task dependence analysis (#2163) (by xumingkuan)
• [misc] Add experimental Python 3.9 support (#2157) (by 彭于斌)
• [lang] Expose SNode ID to python (#2162) (by Ye Kuang)
• [cuda] Add argument "gpu_max_reg" to ti.init (#2161) (by Yuanming Hu)
• [opt] [async] Improve full_simplify and optimize_dead_store (#2160) (by xumingkuan)
• [misc] Enable pypi upload for macOS (#2159) (by Ye Kuang)
• [misc] Deleted 3 debug messages in codegen_cc.cpp (#2158) (by Jiasheng Zhang)
• [sparse] Make memory allocator more robust (#2156) (by Yuanming Hu)
• [misc] Fix SNode max_num_elements to use int64 (#2154) (by Ye Kuang)
• [opt] Simplify bit_cast of bit_cast (#2152) (by xumingkuan)
• [async] [bug] Fix missing memory access options in async mode (#2150) (by xumingkuan)
• [type] Fix struct-for block dim on bit_structs (#2151) (by Yuanming Hu)
• [misc] Add a Github Action workflow to trigger on publishing a release (#2149) (by Ye Kuang)
• [type] Fix arm64 flush to zero (#2148) (by Yuanming Hu)
• [type] Update custom data type APIs (#2147) (by Yuanming Hu)
• [type] Support basic custom int/float types on metal (#2145) (by Ye Kuang)
• [type] Local adder structure (#2136) (by Xuanda Yang)

Changelog:

• [type] Improve GlobalStoreStmt precision on custom floats (#2093) (by Yuanming Hu)
• [type] [bug] Support atomic add negative numbers for custom types (#2091) (by Jiafeng Liu)
• [bug] Fix unsort_edges to use tmp_data (#2092) (by Ye Kuang)
• [misc] Expose "StateFlowGraph::benchmark_rebuild_graph()" (#2090) (by Yuanming Hu)
• [async] Add bit mask in the SFG node edges container (#2089) (by Ye Kuang)
• [async] Use monospace font for DOT graph (#2087) (by xumingkuan)
• [lang] [bug] Clamp int constant to its native range (#2083) (by Ye Kuang)
• [type] Support atomic add on CustomFloatType (#2085) (by Jiafeng Liu)
• [async] Call sort_edges() once during graph rebuilding (#2086) (by Ye Kuang)
• [ir] SNodeOpStmt use GlobalPtrStmt for all SNodeOpTypes (#2082) (by xumingkuan)
• [Lang] Fix ndrange symbol leakage in AST transformer (#2081) (by 彭于斌)
• [async] Add customized container for async state to node map (by Ye Kuang)
• [type] Support atomic add on CustomIntType (#2078) (by Yuanming Hu)
• Revert "clean up (#2070)" (#2079) (by Ye Kuang)
• clean up (#2070) (by Ye Kuang)

Changelog:

• [perf] Fix block local storage transform performance (#2057) (by Yuanming Hu)
• [OpenGL] [bug] Fix OpenGL for global fields with unsigned type (#2051) (by 彭于斌)
• [type] [refactor] Add compute_type for CustomIntType (#2047) (by Jiafeng Liu)
• [Example] Add mass_spring_game.py (#2056) (by Yuanming Hu)
• [Example] Add vortex_rings.py (#2055) (by Yuanming Hu)

Changelog:

• [test] Random nits on tests (#2019) (by 彭于斌)
• [type] Refactor bit pointers for SNodeLookupStmt in type check transform (#2046) (by Xuanda Yang)
• [type] Support SNodeLookUp with bit_array (#2041) (by Xuanda Yang)
• [misc] Update README.md (#2044) (by Yuanming Hu)

Changelog:

• [test] Random nits on tests (#2019) (by 彭于斌)
• [type] Refactor bit pointers for SNodeLookupStmt in type check transform (#2046) (by Xuanda Yang)
• [type] Support SNodeLookUp with bit_array (#2041) (by Xuanda Yang)
• [misc] Update README.md (#2044) (by Yuanming Hu)

Highlights:

• Language and syntax
  • Rename "ti.cache_shared" to "ti.block_local" (#2030) (by Zhiya Luo, welcome!)

Full changelog:

• [opt] Algebraic simplification for sar/shl/shr (#2031) (by xumingkuan)
• [type] Support bit-level read and write in Python-scope (#2029) (by Jiafeng Liu)
• [Lang] [refactor] Rename "ti.cache_shared" to "ti.block_local" (#2030) (by Zhiya Luo)
• [type] Refactor bit pointers (#2028) (by Yuanming Hu)
• [async] Use loop-unique info for fusion (#2012) (by xumingkuan)
• [ir] [opt] Demote BitExtractStmt into a series of binary operations for optimization (#1795) (by 彭于斌)
• [misc] Fix Type* ownership in Python-scope (#2026) (by Yuanming Hu)
• [example] Interpolate vertices for mciso_advanced.py to make it smoother (#1991) (by 彭于斌)
• [opengl] [refactor] Move rand_state from runtime to gtmp to reduce SSBO numbers (#2021) (by 彭于斌)
• [type] Add BitArrayType and corresponding SNodes (#2017) (by Xuanda Yang)
• [opengl] [refactor] Reduce SSBO numbers: merge earg with args (#2020) (by 彭于斌)
• [misc] Add clear_profile_info() (#2018) (by Ye Kuang)
• [opengl] [perf] Grid-stride loop for all type of loops (#2016) (by 彭于斌)
• [metal] Support pointer SNode in codegen (#2015) (by Ye Kuang)
• [async] Support activation demotion in "if" statements (#2009) (by Yuanming Hu)

Changelog:

• [Type] Support bit-level load and store (#1996) (by Jiafeng Liu)
• [sparse] Fix allocator initialization (#2010) (by Yuanming Hu)
• [async] Improve benchmarks (#2005) (by Yuanming Hu)
• [metal] Revise NodeManager's implementation due to weak memory order (#2008) (by Ye Kuang)
• [OpenGL] [perf] Utilize glDispatchComputeIndirect to prevent sync when dynamic ranges are used (#2007) (by 彭于斌)

Highlights:

• Examples
  • Fix euler.py example (#1999) (by Yadi Cao)
• Performance improvements
  • Automatically use read-only data cache load (#1998) (by Yuanming Hu)

Full changelog:

• [metal] Add SNodeRep_pointer (#2000) (by Ye Kuang)
• [misc] Add a link to Chinese video tutorial on Bilibili (#2003) (by Yuanming Hu)
• [async] [bug] Fix a bug in fusion and add some debug info (#2002) (by xumingkuan)
• [Perf] [cuda] Automatically use read-only data cache load (#1998) (by Yuanming Hu)
• [Example] Fix euler.py example (#1999) (by Yadi Cao)
• [metal] Refactor sparse shader impl in prep for pointer SNode (#1994) (by Ye Kuang)
• [error] Improve kernel argument type mismatch error message (#1995) (by Yuanming Hu)