| c76598a | Steven Johnson | 14 April 2022, 23:37:08 UTC | decorator cleanup | 14 April 2022, 23:37:08 UTC |
| 4dd40d0 | Steven Johnson | 14 April 2022, 21:01:53 UTC | Merge branch 'srj/abstract-generator' into srj/pygen-class | 14 April 2022, 21:01:53 UTC |
| 13a87b7 | Steven Johnson | 14 April 2022, 21:01:39 UTC | Merge branch 'main' into srj/abstract-generator | 14 April 2022, 21:01:39 UTC |
| 60a909f | Steven Johnson | 14 April 2022, 16:49:38 UTC | Fix type-mangling for vst on arm32 for LLVM15 (#6705) | 14 April 2022, 16:49:38 UTC |
| 77f7f5e | Steven Johnson | 13 April 2022, 21:31:17 UTC | Python: make Func implicitly convertible to Stage (#6702) (#6704) This allows for `compute_with` and `rfactor` to work more seamlessly in Python. Also: - Move two compute_with() variant bindings from PyFunc and PyStage to PyScheduleMethods, as they are identical between the two - drive-by removal of redundant `py::implicitly_convertible<ImageParam, Func>();` call | 13 April 2022, 21:31:17 UTC |
| 87c0cc9 | Andrew Adams | 12 April 2022, 23:38:56 UTC | llvm no longer wants a type suffix on vst intrinsics (#6701) * llvm no longer wants a type suffix on vst intrinsics * Fix silly mistake * Change 64-bit only Co-authored-by: Andrew Adams <anadams@adobe.com> | 12 April 2022, 23:38:56 UTC |
| 3d7b977 | Steven Johnson | 12 April 2022, 23:33:02 UTC | Drop support for Matlab extensions (#6696) * Drop support for Matlab extensions Anecdotally, this hasn't been used in ~years, and the original author (@dsharletg) had suggested dropping it a while back. I'm going to propose we go ahead and drop it for Halide 15 and see who complains. * Fixes for top-of-tree LLVM * Update force_include_types.cpp * trigger buildbots * Update CodeGen_LLVM.cpp | 12 April 2022, 23:33:02 UTC |
| 4da8932 | Steven Johnson | 12 April 2022, 04:45:12 UTC | Remove deprecated JIT handler setters (#6699) | 12 April 2022, 04:45:12 UTC |
| 009d86f | Steven Johnson | 12 April 2022, 04:45:00 UTC | Remove deprecated versions of Func::prefetch() (#6698) | 12 April 2022, 04:45:00 UTC |
| 3944fb0 | Alexander Root | 12 April 2022, 02:41:52 UTC | Faster `widening_mul(int16x, int16x) -> int32x` for x86 (AVX2 and SSE2) (#6677) * add widening_mul using vpmaddwd for AVX2 * add vpmaddwd/pmaddwd test * add widening_mul with pmaddwd for SSE2 | 12 April 2022, 02:41:52 UTC |
| 08325a4 | Steven Johnson | 11 April 2022, 23:36:16 UTC | Fixes for top-of-tree LLVM (#6697) | 11 April 2022, 23:36:16 UTC |
| c8216fd | Steven Johnson | 11 April 2022, 17:02:32 UTC | Merge branch 'main' into srj/abstract-generator | 11 April 2022, 17:02:32 UTC |
| a9eb446 | Steven Johnson | 11 April 2022, 17:02:29 UTC | Update abstractgeneratortest_generator.cpp | 11 April 2022, 17:02:29 UTC |
| f906eba | Steven Johnson | 11 April 2022, 16:47:39 UTC | Silence "unknown warning" in Clang 13 (#6693) Clang 13 removed the `return-std-move-in-c++11` warning entirely, so specifying it now warns that the warning is unknown. | 11 April 2022, 16:47:39 UTC |
| d568469 | Steven Johnson | 11 April 2022, 16:40:27 UTC | Add `break` to avoid 'possible unintentional fallthru' warning (#6694) | 11 April 2022, 16:40:27 UTC |
| 54f3977 | Steven Johnson | 08 April 2022, 23:25:34 UTC | Always mark _ucon as 'unused' in Codegen_C (#6691) * Always mark _ucon as 'unused' in Codegen_C, even if asserts are enabled, since generated closure functions may not use it * halide_unused -> halide_maybe_unused * fix test_internal * More halide_unused -> halide_maybe_unused | 08 April 2022, 23:25:34 UTC |
| b260096 | Steven Johnson | 08 April 2022, 23:02:02 UTC | Merge branch 'main' into srj/abstract-generator | 08 April 2022, 23:02:02 UTC |
| 887d340 | Steven Johnson | 08 April 2022, 19:01:27 UTC | Upgrade to clang-format 13 (#6689) Goal here: eliminate the need for a local version of llvm/clang-12, and don't stay too far behind the toolchain. As always, clang-format doesn't promise backwards compatibility, but the main differences in formatting are: - more regularization of spaces at the start of comments (I like this change) - minor difference of formatting of function-pointer-type declarations (not a fan of this, but I can't find a way to disable it and it's only really used in a handful of place in the Python bindings) | 08 April 2022, 19:01:27 UTC |
| b5840f7 | Steven Johnson | 08 April 2022, 00:59:14 UTC | Drop support for LLVM12 (#6686) * Drop support for LLVM12 Halide 15 only needs to support LLVM13 and LLVM13. Drop all the special-casing for LLVM12. * Update packaging.yml * Update presubmit.yml * 13 * more * Update presubmit.yml * woo * Update presubmit.yml * Update run-clang-tidy.sh * Update run-clang-tidy.sh * Update .clang-tidy * Update .clang-tidy * wer * Update Random.cpp * wer * sdf * sdf * Update packaging.yml | 08 April 2022, 00:59:14 UTC |
| e549be7 | Steven Johnson | 08 April 2022, 00:03:34 UTC | Remove deprecated `build()` support from Generators (#6684) This was deprecated in Halide 14; let's remove it entirely for Halide 15. | 08 April 2022, 00:03:34 UTC |
| f64bd08 | Steven Johnson | 07 April 2022, 23:59:49 UTC | Remove deprecated `Halide::Output` type (#6685) It was deprecated (in favor of `OutputFileType` in Halide 14; let's remove it entirely for Halide 15. | 07 April 2022, 23:59:49 UTC |
| fe96aaa | Steven Johnson | 07 April 2022, 21:30:08 UTC | Fix "set but not used" warnings/errors (#6683) * Fix "set but not used" warnings/errors Apparently XCode 13.3 has smarter warnings about unused code and emits warnings/errors for these, so let's clean them up. * Also fix missing `ssize_t` usage | 07 April 2022, 21:30:08 UTC |
| 1ed5cd3 | Steven Johnson | 07 April 2022, 18:37:18 UTC | WIP | 07 April 2022, 18:37:18 UTC |
| bcd1ed2 | Steven Johnson | 07 April 2022, 17:10:15 UTC | Update Makefile | 07 April 2022, 17:10:15 UTC |
| d18fe26 | Steven Johnson | 07 April 2022, 17:01:54 UTC | Merge branch 'srj/abstract-generator' into srj/pygen-class | 07 April 2022, 17:01:54 UTC |
| b14e0df | Steven Johnson | 07 April 2022, 17:01:48 UTC | Merge branch 'main' into srj/pygen-class | 07 April 2022, 17:01:48 UTC |
| 5edba1a | Steven Johnson | 07 April 2022, 17:01:34 UTC | Merge branch 'main' into srj/abstract-generator | 07 April 2022, 17:01:34 UTC |
| 6cca294 | Steven Johnson | 07 April 2022, 01:24:13 UTC | WIP | 07 April 2022, 01:24:13 UTC |
| 1d1b556 | Steven Johnson | 06 April 2022, 23:25:04 UTC | Halide::Tools::save_image() should accept buffers with `const` types (#6679) | 06 April 2022, 23:25:04 UTC |
| 1ff65e5 | Steven Johnson | 06 April 2022, 18:09:16 UTC | Update PyGenerator.cpp | 06 April 2022, 18:09:16 UTC |
| a9e03e6 | Steven Johnson | 06 April 2022, 18:03:00 UTC | Merge branch 'srj/abstract-generator' into srj/pygen-class | 06 April 2022, 18:03:00 UTC |
| da9505f | Steven Johnson | 06 April 2022, 18:02:42 UTC | fix pystub | 06 April 2022, 18:02:42 UTC |
| f1538d5 | Steven Johnson | 06 April 2022, 18:02:25 UTC | Merge branch 'srj/abstract-generator' into srj/pygen-class | 06 April 2022, 18:02:25 UTC |
| c9d7b5a | Steven Johnson | 06 April 2022, 17:54:19 UTC | Merge branch 'main' into srj/abstract-generator | 06 April 2022, 17:54:19 UTC |
| ad0408e | Steven Johnson | 06 April 2022, 17:52:14 UTC | Clean up Python extensions in python_bindings (#6670) * Remove the nobuild/partialbuildmethod tests from python_bindings/ They no longer serve a purpose and are redundant to other tests. * WIP * Update pystub.py * wip * wip * wip * Update TargetExportScript.cmake * Update PythonExtensionHelpers.cmake * PyExtensionGen didn't handle zero-dimensional buffers | 06 April 2022, 17:52:14 UTC |
| 12270a5 | Alex Reinking | 06 April 2022, 16:57:39 UTC | Bump development Halide version to 15.0.0 (#6678) * Bump development Halide version to 15.0.0 * trigger buildbots | 06 April 2022, 16:57:39 UTC |
| 72ad2e6 | Roman Lebedev | 06 April 2022, 16:13:45 UTC | `-mtune=native` CPU autodetection for AMD Zen 3 CPU (#6648) * `-mtune=native` CPU autodetection for AMD Zen 3 CPU * Address review notes. * Fix MSVC build * Address review notes | 06 April 2022, 16:13:45 UTC |
| 9866df2 | Steven Johnson | 05 April 2022, 23:23:25 UTC | Fix ctors for Realization (#6675) For vector-of-Buffers, the ctor took a non-const ref to the argument, which was weird and nonsensical. Replaced with a const-ref version and and an rvalue-ref version; it turns out that literally *all* of the internal calls were able to use the latter, trivially saving some copies. | 05 April 2022, 23:23:25 UTC |
| fdd6500 | Roman Lebedev | 05 April 2022, 20:51:31 UTC | Future-proof 'processor` to `tune processor` (#6673) | 05 April 2022, 20:51:31 UTC |
| 43af5b6 | Steven Johnson | 05 April 2022, 20:38:51 UTC | Allow PyPipeline and PyFunc to realize() scalar buffers (#6674) | 05 April 2022, 20:38:51 UTC |
| 1873afb | Steven Johnson | 05 April 2022, 00:56:03 UTC | Add OutputScalar | 05 April 2022, 00:56:03 UTC |
| f56614e | Shoaib Kamil | 01 April 2022, 14:27:50 UTC | Fix GPU depredication/scalarization (#6669) * Scalarize predicated Loads * Cleanup * Fix gpu_vectorize scalarization for D3D12 * Fix OpenCL scalarization * Minor fixes * Formatting * Address review comments * Move Shuffle impl to CodeGen_GPU_C class * Extra space removal Co-authored-by: Shoaib Kamil <kamil@adobe.com> | 01 April 2022, 14:27:50 UTC |
| 40f895d | Roman Lebedev | 31 March 2022, 22:23:26 UTC | `-mtune=`/`-mcpu=` support for x86 AMD CPU's (#6655) * `-mtune=`/`-mcpu=` support for x86 AMD CPU's * Move processor tune into it's own enum, out of features * clang-format * Target: make Processor more optional * Processor: add explanatory comments which CPU is what * Drop outdated changes * Make comments in Processor more readable / fix BtVer2 comment * Target: don't require passing Processor * Make processor more optional in the features string serialization/verification * Address review notes * Undo introduction of halide_target_processor_t * Fix year for btver2/jaguar | 31 March 2022, 22:23:26 UTC |
| 6b9ed2a | Steven Johnson | 30 March 2022, 22:30:42 UTC | Remove the nobuild/partialbuildmethod tests from python_bindings/ (#6668) They no longer serve a purpose and are redundant to other tests. | 30 March 2022, 22:30:42 UTC |
| 5d2abd3 | Steven Johnson | 29 March 2022, 23:33:42 UTC | Add ldscript code for Python extensions in CMake (#6665) * Add ldscript code for Python extensions in CMake We added ldscripts to the Makefile for Python extensions (to restrict exported symbols to just the PyInit_foo symbol), but neglected to do so for CMake. This corrects that. | 29 March 2022, 23:33:42 UTC |
| 1c6506b | Steven Johnson | 29 March 2022, 23:01:16 UTC | Update readme.md | 29 March 2022, 23:01:16 UTC |
| ff3f203 | Steven Johnson | 29 March 2022, 22:31:46 UTC | WIP | 29 March 2022, 22:31:46 UTC |
| 18ed748 | Steven Johnson | 29 March 2022, 19:15:00 UTC | Merge branch 'main' into srj/abstract-generator | 29 March 2022, 19:15:00 UTC |
| 8aba364 | Steven Johnson | 29 March 2022, 18:56:33 UTC | Allow `make test_apps` to work with ASAN (#6659) * Allow `make test_apps` to work with ASAN With asan or tsan in the target, there is a space in the OPTIMIZE var, so it needs to be quoted. * tickle buildbots * tickle buildbots * tickle buildbots | 29 March 2022, 18:56:33 UTC |
| a4c70fa | Steven Johnson | 29 March 2022, 03:52:20 UTC | clang-format | 29 March 2022, 03:52:20 UTC |
| 251b9df | Steven Johnson | 29 March 2022, 02:08:25 UTC | Various cleanups of AbstractGenerator | 29 March 2022, 02:08:25 UTC |
| ed3f4a7 | Steven Johnson | 29 March 2022, 00:10:49 UTC | Add optional runtime H::R::Buffer access checks (#6660) * Add optional runtime H::R::Buffer access checks This adds some optional `assert()` checks to HRB's `operator()` and friends. They are only enabled if `HALIDE_RUNTIME_BUFFER_CHECK_INDICES=1` is defined at compile time. Also fixes errors found by enabling these assertions and running tests. * Update fast_pow.cpp * clang-format * tickle buildbots * tickle buildbots | 29 March 2022, 00:10:49 UTC |
| c2bebe2 | Steven Johnson | 29 March 2022, 00:10:37 UTC | Python Bindings: fix Python `bool` -> `Expr` implicit conversion (#6657) * Python Bindings: fix Python `bool` -> `Expr` implicit conversion It was implicitly converting to a Halide `Int(32)` literal of value 1, but we want it to match Halide's boolean type of `UInt(1)` * Update basics.py * wip * tickle buildbots | 29 March 2022, 00:10:37 UTC |
| f62426a | Steven Johnson | 28 March 2022, 17:55:21 UTC | Merge branch 'main' into srj/abstract-generator | 28 March 2022, 17:55:21 UTC |
| 14d89f3 | Steven Johnson | 25 March 2022, 23:38:46 UTC | Fix 'variable set but not used` warning/error (#6658) Yes, some compilers complain about this. | 25 March 2022, 23:38:46 UTC |
| 5907f82 | Steven Johnson | 23 March 2022, 18:33:23 UTC | IOKind -> ArgInfoKind | 23 March 2022, 18:33:23 UTC |
| 1350142 | Steven Johnson | 23 March 2022, 18:28:24 UTC | Merge branch 'main' into srj/abstract-generator | 23 March 2022, 18:28:24 UTC |
| 17b537c | Zalman Stern | 22 March 2022, 19:04:53 UTC | Timer based profiler (#6642) * Add support for timer interrupt based profiling, which is useful for bring up on embedded ("bare metal") systems that may not have a full OS with threads. * Update runtime_api file with new routines. * Turn locking back on in timer based profiling case as it can be used in multiprocessor situations. (Both as an option on systems where threads would be fine and on embedded systems which don't have time shared threads but cores are dedicated to the Halide threadpool.) * Add target flag for timer profiling and extend performance_profiler test to cover timer profiling. | 22 March 2022, 19:04:53 UTC |
| 650554a | Steven Johnson | 22 March 2022, 00:31:24 UTC | Eliminate some unnecessary clamping in ClampUnsafeAccesses (#6297) (#6654) * Eliminate some unnecessary clamping in ClampUnsafeAccesses (#6297) * Update ClampUnsafeAccesses.cpp * Update ClampUnsafeAccesses.cpp * Update ClampUnsafeAccesses.cpp | 22 March 2022, 00:31:24 UTC |
| 49db215 | Steven Johnson | 18 March 2022, 19:36:17 UTC | Fix apparent type in PR #6294 (#6653) | 18 March 2022, 19:36:17 UTC |
| 8b87c80 | Steven Johnson | 16 March 2022, 17:01:41 UTC | Merge branch 'main' into srj/abstract-generator | 16 March 2022, 17:01:41 UTC |
| 9ab3566 | Roman Lebedev | 16 March 2022, 04:56:40 UTC | [CMake] Deduplicate `Halide_LLVM_VERSION` and `LLVM_PACKAGE_VERSION` (#6646) | 16 March 2022, 04:56:40 UTC |
| a9ade82 | Steven Johnson | 15 March 2022, 23:40:23 UTC | Merge branch 'main' into srj/abstract-generator | 15 March 2022, 23:40:23 UTC |
| b608583 | Steven Johnson | 15 March 2022, 22:07:46 UTC | Update initialization of WABT `store` field to work with top-of-tree (#6649) The copy and move assignment operators for Store are going away; initialize ours in the | 15 March 2022, 22:07:46 UTC |
| 07dddb7 | Steven Johnson | 10 March 2022, 18:57:29 UTC | Allow profiler feature under wasm iff wasm_threads is enabled (#6643) The profiler requires threads, but works fine when wasm_threads are enabled. | 10 March 2022, 18:57:29 UTC |
| f6628aa | Steven Johnson | 10 March 2022, 18:50:25 UTC | Fix UB in hannk FillWithRandom operation. (#6645) A recent change to libc++ rejects std::uniform_int_distribution<uint8_t> explicitly (https://reviews.llvm.org/D114920). This change prevents this file to build with upcoming revisions of the C++ toolchain. | 10 March 2022, 18:50:25 UTC |
| 105f7e5 | Steven Johnson | 07 March 2022, 20:59:04 UTC | Fix const-correctness in C/C++ backend (Issue #6636) (#6638) in the Load handler, we need to emit the cast in the form `TYPE const *` rather than `const TYPE *`, as TYPE could be `void *`, and the const would bind in a way we don't want. | 07 March 2022, 20:59:04 UTC |
| 5f37d50 | Steven Johnson | 07 March 2022, 20:58:12 UTC | Convert most remaining Generators to prefer statically-dimensioned Inputs and Output where possible (#6641) This is the same as #6620, except that it omits autoschedulers/adams2019/cost_model_generator.cpp (which is unusually complex and not yet settled as to whether the changes are welcome). Basically an attempt to land the uncontroversial parts. | 07 March 2022, 20:58:12 UTC |
| a55ae55 | Alexander Root | 06 March 2022, 20:21:32 UTC | Clear bounds info on casts when value bounds are undefined (for overflow types) (#6640) | 06 March 2022, 20:21:32 UTC |
| 979e204 | Karthik Nishanth | 06 March 2022, 00:25:10 UTC | python_bindings: acquire GIL before printing (#6635) | 06 March 2022, 00:25:10 UTC |
| 0b8e263 | Steven Johnson | 06 March 2022, 00:24:29 UTC | Clean up python_binding Makefile (#6634) * Clean up python_binding Makefile Some of the tests with Generator dependencies had deps set up in a weird way that made some downstream work painful. Cleaned up. Also vastly reduced build-time noise. * Update Makefile * Add linker scripts for Python extensions | 06 March 2022, 00:24:29 UTC |
| 52b9640 | Steven Johnson | 04 March 2022, 19:26:11 UTC | Update Generator.cpp | 04 March 2022, 19:26:11 UTC |
| 76ce1d8 | Steven Johnson | 04 March 2022, 18:36:42 UTC | clang-format | 04 March 2022, 18:36:42 UTC |
| d4b41d1 | Steven Johnson | 04 March 2022, 17:54:15 UTC | Update AbstractGenerator.h | 04 March 2022, 17:54:15 UTC |
| e651258 | Steven Johnson | 04 March 2022, 01:29:24 UTC | Merge branch 'main' into srj/abstract-generator | 04 March 2022, 01:29:24 UTC |
| 3827279 | Steven Johnson | 04 March 2022, 00:49:18 UTC | Python Bindings didn't allow for zero-D Funcs, ImageParams, Buffers (#6633) * Python Bindings didn't allow for zero-D Funcs, ImageParams, Buffers There were no overloads or tests for accessing the element of any of these in the zero-D case, and the obvious syntax (`[]`, to mirror C++ `()` in these cases) isn't legal in Python. To support this uncommon-but-necessary case, I'm proposing that we use the syntax `[None]`, which isn't pretty, but is less bad than other options I've considered so far. (Suggestions welcome.) * Use [()] instead of [None] | 04 March 2022, 00:49:18 UTC |
| 6933e0e | Steven Johnson | 03 March 2022, 01:07:36 UTC | AbstractGenerator (rebased, v3) | 03 March 2022, 01:07:36 UTC |
| 86728d7 | Alexander Root | 26 February 2022, 18:56:09 UTC | Wild match object is not foldable (#6623) | 26 February 2022, 18:56:09 UTC |
| caa3ad0 | Volodymyr Kysenko | 24 February 2022, 01:09:55 UTC | Avoid double narrowing in widening_add/widening_sub if type is 8-bit (#6629) | 24 February 2022, 01:09:55 UTC |
| b2edbc9 | Alexander Root | 22 February 2022, 20:09:23 UTC | Disallow `Type::narrow()` and `Type::widen()` from producing bitwidths between 1 and 8 bits (#6622) * Disallow Type::narrow() and Type::widen() from producing bitwidths between 1 and 8 bits * Narrowing a 1-bit type should error | 22 February 2022, 20:09:23 UTC |
| a4ed033 | Alex Reinking | 22 February 2022, 04:31:55 UTC | Make IRComparer consider nans to be less than non-nans. (#6626) * Make IRComparer consider nans to be less than non-nans. Fixes #6624 * Fix test * Respond to reviewer comments. * Return -1 instead of +1 | 22 February 2022, 04:31:55 UTC |
| 16bfa2f | Alexander Root | 21 February 2022, 18:42:15 UTC | remove incorrect docs on widening_add (#6625) | 21 February 2022, 18:42:15 UTC |
| be1269b | Andrew Adams | 17 February 2022, 21:44:35 UTC | Add Stage::unscheduled() | 18 February 2022, 22:26:15 UTC |
| 0786dd4 | Andrew Adams | 02 February 2022, 18:36:57 UTC | Fix atomics test | 18 February 2022, 22:26:15 UTC |
| 6424aaa | Andrew Adams | 01 February 2022, 22:24:11 UTC | Reenable warning about unscheduled update definitions and fix associated issues in the tests and apps. This is an old warning that stopped triggering because it wasn't tested. We should either remove it, fix the trigger conditions, or perhaps make it an error. This PR fixes the trigger conditions and fixes all instances of the warning in our tests and apps. The warning triggers if you schedule some but not all of the update definitions of a Func. It's to protect against the common error of only scheduling the pure definition of something like a summation. The warning can be suppressed by inserting a call to func.update(idx). | 18 February 2022, 22:26:15 UTC |
| 7373eb9 | Steven Johnson | 17 February 2022, 17:07:46 UTC | Move GeneratorContext into a standalone class (#6618) * Move GeneratorContext into a standalone class * Minor Fixes * clang-tidy * Update Generator.cpp * Update Generator.cpp | 17 February 2022, 17:07:46 UTC |
| 846592f | Steven Johnson | 16 February 2022, 22:41:18 UTC | Update WABT version to the just-released 1.027 (instead of main) (#6619) * Update WABT version to the just-released 1.027 (instead of main) * tickle buildbots | 16 February 2022, 22:41:18 UTC |
| 4ccd0ec | Alex Reinking | 16 February 2022, 01:25:09 UTC | Remove halide_config.cmake from Makefile build. Fixes #6615 (#6616) | 16 February 2022, 01:25:09 UTC |
| f9189dc | Steven Johnson | 16 February 2022, 01:24:57 UTC | Update apps/hannk to use TFLite 2.8.0 (#6617) | 16 February 2022, 01:24:57 UTC |
| 3628b67 | Steven Johnson | 15 February 2022, 00:30:25 UTC | Minor Generator cleanup (#6613) | 15 February 2022, 00:30:25 UTC |
| ba86c2e | Steven Johnson | 15 February 2022, 00:29:13 UTC | Unbreak WABT again by using main instead of a commit (#6614) | 15 February 2022, 00:29:13 UTC |
| 38032e8 | Basile Clement | 14 February 2022, 18:43:50 UTC | Only commutative reductions can be parallelized (#6609) Because parallelization changes the order of computation within the reduction, parallelizing associative but non-commutative reductions can result in (non-deterministically) incorrect results in the same way `reorder`ing them can. For instance Halide currently accepts the following code, but generates non-deterministic outputs on GPU. On CPU with `.parallel(r.x)`, OpenMP rejects the generated code (correctly) stating that the `#pragma omp atomic` is invalid for the same reasons. ```c++ #include <stdio.h> #include "Halide.h" using namespace Halide; int main(int argc, char **argv) { Halide::Func A("A"), B("B"); Halide::Var i("i"); A(i) = i; B() = -1; Halide::RDom r(0, 1024); B() = A(r.x); A.compute_root(); B.update().atomic().gpu_blocks(r.x); B.compile_jit(get_host_target().with_feature(Target::CUDA)); Halide::Buffer<int32_t> b = B.realize(); printf("%d\n", b()); return 0; } ``` | 14 February 2022, 18:43:50 UTC |
| 832efeb | Steven Johnson | 11 February 2022, 22:46:23 UTC | Update WasmExecutor for WABT API changes (#6612) * Update WasmExecutor for WABT API changes The API for WABT has changed in top-of-tree. This fixes the incompatibilities. It also (temporarily) changes to pulling WABT from top-of-tree (rather than a known release) as the most recent labeled release doesn't have these changes. (We'll change back to a fixed WABT release when the next one is released.) * Update CMakeLists.txt | 11 February 2022, 22:46:23 UTC |
| dd32beb | Zalman Stern | 11 February 2022, 15:53:50 UTC | Fixes to support LLVM with opaque pointers. (#6608) Removes uses of LLVM API routines that recover pointed to types from LLVM pointer types as that functionality is going away with opaque pointers. | 11 February 2022, 15:53:50 UTC |
| 340f8f2 | Andrew Adams | 08 February 2022, 21:55:58 UTC | Add comment | 10 February 2022, 20:21:19 UTC |
| 5993334 | Andrew Adams | 08 February 2022, 19:43:51 UTC | Fix bug in mul_shift_right matching widening_mul handles mixed-signedness args, but mul_shift_right doesn't. We shouldn't map a mixed-signedness widening_mul to a mul_shift_right. | 10 February 2022, 20:21:19 UTC |
| f28c07a | Steven Johnson | 02 February 2022, 21:52:02 UTC | Allow building with LLVM15 (#6603) | 02 February 2022, 21:52:02 UTC |
| 1e7040c | Steven Johnson | 31 January 2022, 21:30:55 UTC | Convert Buffer<> usage in python_bindings/ to use static dimensions (#6591) * Various fixes to static-dimensioned Buffer - Buffer needed to make some methods constexpr, and also had a broken return value for `as<>()` - Various templated methods needed to add `int Dims` as a parameter - Generator::add_input() and add_output() needed specializations for static-buffer types - sliced() and embedded() should use the default values for InClassDimStorage - halide_image_io should use a named constant * Convert Buffer<> usage in python_bindings/ to use static dimensions ...where it seems sensible to do so, that is. * Update Generator.h | 31 January 2022, 21:30:55 UTC |
| b9a4cb3 | Steven Johnson | 31 January 2022, 21:30:40 UTC | Convert Buffer<> usage in test/generators to use static dimensions (#6592) * Various fixes to static-dimensioned Buffer - Buffer needed to make some methods constexpr, and also had a broken return value for `as<>()` - Various templated methods needed to add `int Dims` as a parameter - Generator::add_input() and add_output() needed specializations for static-buffer types - sliced() and embedded() should use the default values for InClassDimStorage - halide_image_io should use a named constant * Convert Buffer<> usage in test/generators to use static dimensions ...where it seems sensible to do so, that is. (also test/integration since it's small) * Update Generator.h | 31 January 2022, 21:30:40 UTC |
| d147a68 | Steven Johnson | 27 January 2022, 20:29:13 UTC | Rename BufferDimsUnconstrained -> AnyDims (#6594) Existing name is kinda clunky, this seems terser and cleaner | 27 January 2022, 20:29:13 UTC |
