821cfed | Patricia Suriana | 12 January 2018, 22:12:13 UTC | Merge branch 'master' of github.com:halide/Halide into unbounded_bugs | 12 January 2018, 22:12:13 UTC |
95c8e56 | Patricia Suriana | 12 January 2018, 22:11:20 UTC | Merge pull request #2691 from halide/compute_with_inlined Fix bugs in compute_with when the fused groups refer to inlined functions | 12 January 2018, 22:11:20 UTC |
5110819 | Patricia Suriana | 12 January 2018, 20:08:09 UTC | Use std::reverse | 12 January 2018, 20:08:09 UTC |
8f4de18 | Patricia Suriana | 12 January 2018, 17:51:49 UTC | Fix bugs in compute_with when the fused groups refer to inlined functions | 12 January 2018, 17:51:49 UTC |
2994035 | Dillon Sharlet | 10 January 2018, 18:16:01 UTC | Merge pull request #2683 from halide/zvi-avx2 X86: AVX2 improvements in codegen and runtime functions (by ZviRackover) | 10 January 2018, 18:16:01 UTC |
ec9f0d9 | dsharletg | 10 January 2018, 18:15:25 UTC | Remove redundant static. | 10 January 2018, 18:15:25 UTC |
834a007 | Zalman Stern | 08 January 2018, 21:14:42 UTC | Merge pull request #2684 from halide/remove_spurious_llvm_dir Remove bad llvm dir commit. | 08 January 2018, 21:14:42 UTC |
2a94f9e | Z Stern | 08 January 2018, 21:08:43 UTC | Remove bad llvm dir commit. | 08 January 2018, 21:08:43 UTC |
84825c9 | Zalman Stern | 08 January 2018, 20:46:18 UTC | Merge pull request #2657 from halide/thread_safe_jit_call Initial support for passing per-call params to realize et al. | 08 January 2018, 20:46:18 UTC |
e35ba79 | dsharletg | 08 January 2018, 19:08:23 UTC | Merge branch 'avx2-intrinsics' of https://github.com/ZviRackover/Halide into zvi-avx2 | 08 January 2018, 19:08:23 UTC |
9fd8337 | Andrew Adams | 08 January 2018, 16:31:39 UTC | Merge pull request #2682 from halide/mingw_libz Fix llvm trunk + mingw | 08 January 2018, 16:31:39 UTC |
4407a69 | Andrew Adams | 08 January 2018, 16:29:59 UTC | Fix llvm trunk + mingw | 08 January 2018, 16:29:59 UTC |
4e8c760 | Z Stern | 06 January 2018, 02:04:03 UTC | Merge branch 'thread_safe_jit_call' of https://github.com/halide/Halide into thread_safe_jit_call | 06 January 2018, 02:04:03 UTC |
b5fc584 | Z Stern | 06 January 2018, 02:03:31 UTC | Address review feedback. | 06 January 2018, 02:03:31 UTC |
88210f6 | Z Stern | 05 January 2018, 23:04:51 UTC | Merge branch 'master' into thread_safe_jit_call | 05 January 2018, 23:04:51 UTC |
c62084f | Zvi Rackover | 05 January 2018, 15:41:35 UTC | X86 Runtime: Use AVX2 for pmaddwd | 05 January 2018, 21:59:23 UTC |
3fd228c | Zvi Rackover | 05 January 2018, 08:23:14 UTC | X86: Implement abs builtin functions with target-independent LLVM IR LLVM's X86 backend is capable of lowering the canonical form of abs to the pabs instructions. | 05 January 2018, 21:58:37 UTC |
d20594a | Zvi Rackover | 04 January 2018, 20:14:17 UTC | X86: Improve codegen for AVX2 Add missing patterns and runtime functions. Strengthen tests for cases that were already supported. | 05 January 2018, 21:49:19 UTC |
8c23a19 | Zalman Stern | 05 January 2018, 21:21:24 UTC | Merge pull request #2676 from ZviRackover/avx2-saturated-truncate-3 X86 Runtime: AVX2 implementation of saturated truncate functions | 05 January 2018, 21:21:24 UTC |
870937e | Steven Johnson | 05 January 2018, 17:41:39 UTC | Merge pull request #2677 from halide/srj-buf Allow Buffer::copy() to work to/from Buffer<void> | 05 January 2018, 17:41:39 UTC |
7b1f310 | Steven Johnson | 04 January 2018, 20:49:32 UTC | make_with_shape_of(): only use src.type() when T_is_void | 04 January 2018, 20:49:32 UTC |
d7c148e | Z Stern | 04 January 2018, 20:29:36 UTC | Merge branch 'master' into thread_safe_jit_call | 04 January 2018, 20:29:36 UTC |
23768d9 | Steven Johnson | 04 January 2018, 19:59:38 UTC | Allow Buffer::copy() to work to/from Buffer<void> make_with_shape_of() required the src and dst to be strongly typed, but there's no reason to do so; being able to copy to/from a Buffer<void> is fine as long as runtime types match. Also, did unrelated drive-by fix in Bounds.h to fix an int-vs-size_t compiler warning for some configurations. | 04 January 2018, 19:59:38 UTC |
0f8be63 | Patricia Suriana | 04 January 2018, 16:13:54 UTC | Merge branch 'master' of github.com:halide/Halide into unbounded_bugs | 04 January 2018, 16:24:02 UTC |
7ae1b37 | Zvi Rackover | 04 January 2018, 06:51:58 UTC | X86 Test: Fix check with missing AVX2 predicate | 04 January 2018, 06:51:58 UTC |
5ac08fc | Zvi Rackover | 04 January 2018, 05:52:17 UTC | X86 Runtime: Fix pattern to select AVX2 intrinsic only when target feature is enabled. | 04 January 2018, 05:52:17 UTC |
90a0e2d | Steven Johnson | 04 January 2018, 03:14:08 UTC | Merge pull request #2674 from halide/srj-crop Fix correctness/device_crop for HVX targets | 04 January 2018, 03:14:08 UTC |
f58ec44 | Zvi Rackover | 04 January 2018, 01:05:39 UTC | X86 Runtime: AVX2 implementation of saturated truncate functions Create the AVX2 runtime container and add implementations of saturated truncate functions. | 04 January 2018, 02:55:11 UTC |
93fc372 | Steven Johnson | 04 January 2018, 00:23:30 UTC | Fix correctness/device_crop for HVX targets | 04 January 2018, 00:23:30 UTC |
4b9cf8e | Steven Johnson | 04 January 2018, 00:17:38 UTC | Merge pull request #2665 from halide/srj-pyparam Allow Param<void> | 04 January 2018, 00:17:38 UTC |
88236d5 | Steven Johnson | 03 January 2018, 22:15:08 UTC | Merge branch 'master' into srj-pyparam | 03 January 2018, 22:15:08 UTC |
51b7657 | Andrew Adams | 03 January 2018, 20:47:31 UTC | Merge pull request #2673 from halide/srj-llvm7 Update build files to work with LLVM7 | 03 January 2018, 20:47:31 UTC |
8657c93 | Steven Johnson | 03 January 2018, 19:33:19 UTC | Update build files to work with LLVM7 | 03 January 2018, 19:33:19 UTC |
18b2829 | ronlieb | 03 January 2018, 18:48:03 UTC | Merge pull request #2658 from halide/no-assume-hvx Refactor apps building Hexagon | 03 January 2018, 18:48:03 UTC |
23ecab2 | Steven Johnson | 03 January 2018, 18:47:09 UTC | Add copyctor and operator= overloads for Param | 03 January 2018, 18:47:09 UTC |
e6b92c4 | Steven Johnson | 03 January 2018, 18:09:38 UTC | Merge branch 'master' into srj-pyparam | 03 January 2018, 18:09:38 UTC |
3c02a8f | Dillon Sharlet | 03 January 2018, 07:11:03 UTC | Merge pull request #2670 from halide/elide_nop_hexagon_offload_debug_output Attempt to elide Hexagon offload passes if they are not enabled. | 03 January 2018, 07:11:03 UTC |
2a6f4bc | Steven Johnson | 03 January 2018, 00:48:58 UTC | Tweak IsRoundtrippable to work with Param<const char *> | 03 January 2018, 00:48:58 UTC |
3c88e8b | Zalman Stern | 02 January 2018, 23:45:18 UTC | Merge pull request #2671 from halide/hvx_device_crop HVX device crop | 02 January 2018, 23:45:18 UTC |
b4ffbbc | Steven Johnson | 02 January 2018, 22:43:51 UTC | Smarten Param::set() to allow existing API to work well | 02 January 2018, 22:43:51 UTC |
bae9407 | Z Stern | 02 January 2018, 20:49:46 UTC | Add test coverage for Hexagon device cropping. | 02 January 2018, 20:49:46 UTC |
505a8b3 | Z Stern | 02 January 2018, 20:37:27 UTC | Implement device_crop for Hexagon remote device API. | 02 January 2018, 20:37:27 UTC |
df43a78 | Steven Johnson | 02 January 2018, 18:46:15 UTC | Add explicit types to Param::set() where needed | 02 January 2018, 18:46:15 UTC |
41738d9 | Steven Johnson | 28 December 2017, 23:00:17 UTC | Add arg names to Param ctors | 28 December 2017, 23:00:17 UTC |
3e47435 | Z Stern | 26 December 2017, 14:15:22 UTC | Attempt to elide Hexagon offload passes if they are not enabled. Mostly to eliminate superfluous debug output which has been annoying me for a fair while. | 26 December 2017, 14:15:22 UTC |
85efa34 | Steven Johnson | 23 December 2017, 02:42:40 UTC | Allow Param<void> Disclaimer: the goal of this is really to simplify the Python Binding implementation, so it may be misguided. That said, the change is pretty modest and safe, I think. The problem: in the Python bindings, there is no equivalent to a compile-time type bindng; for Param, the type is always specified by a Halide::Type, rather than a compile-time type. Since the existing Param structure *requires* a compile-time type, the existing Python bindings jump through great hoops: it provides a set of bindings for every likely instantiation of Param<T>, then provides an extremely complex set of factory methods to create the right underlying wrapper based on the Type (and other arguments) you pass. Since Param<> already contains dynamic type information (by way of Parameter::type()), this just expands on it (much as Buffer<void> vs Buffer<int> does): Param<void> allows the type to be specified at runtime (rather than compiletime), then does typechecking on the runtime type. The one exception, of course, is the get() and set() method, which still need to be special-cased; however, this is *much* less onerous. Realistically, no 'real' Halide code is ever likely to make use of Param<void>, but it does make binding to dynamic languages like Python substantially simpler. (Note that Param<>::address() was totally unused, and so it was removed.) | 23 December 2017, 02:42:40 UTC |
d2fa706 | Dillon Sharlet | 22 December 2017, 21:13:40 UTC | Merge pull request #2596 from darkbuck/master Enable OpenCL half support | 22 December 2017, 21:13:40 UTC |
446d070 | Steven Johnson | 22 December 2017, 18:25:32 UTC | Merge pull request #2662 from halide/srj-pyns Major namespace refactoring of python_bindings | 22 December 2017, 18:25:32 UTC |
bea81d9 | Michael Liao | 04 December 2017, 04:21:40 UTC | Enable OpenCL half support - Add a new 'cl_half' target feature to enable OCL half support. - Define math functions for f16. - Fix a minor issue in is_nan implementation on half. - Enable python support as well. | 22 December 2017, 00:35:23 UTC |
40720ba | Steven Johnson | 21 December 2017, 22:28:52 UTC | Major namespace refactoring of python_bindings | 22 December 2017, 00:19:19 UTC |
a213b50 | Steven Johnson | 21 December 2017, 23:29:05 UTC | Merge pull request #2661 from halide/srj-pyfix Fix Python 2.7 incompatibility in Python Bindings | 21 December 2017, 23:29:05 UTC |
1a8af34 | Steven Johnson | 21 December 2017, 23:21:20 UTC | Fix Python 2.7 incompatibility in Python Bindings Boost.Python has different constraints for Py2 vs Py3. | 21 December 2017, 23:21:20 UTC |
85de775 | Steven Johnson | 21 December 2017, 22:02:23 UTC | Merge pull request #2654 from halide/srj-pyvar Bring Python bindings for Halide::Var 100% up to date, with complete … | 21 December 2017, 22:02:23 UTC |
52cd4b8 | Steven Johnson | 21 December 2017, 22:02:14 UTC | Merge pull request #2653 from halide/srj-pytype Update Python bindings for Type | 21 December 2017, 22:02:14 UTC |
e28b3ea | Steven Johnson | 21 December 2017, 21:59:17 UTC | Merge pull request #2652 from halide/srj-pytarget Update Python bindings for Target | 21 December 2017, 21:59:17 UTC |
2240709 | Steven Johnson | 21 December 2017, 21:58:28 UTC | use 'py' as alias for 'boost::python' | 21 December 2017, 21:58:28 UTC |
3c904c2 | Steven Johnson | 21 December 2017, 21:55:17 UTC | Use 'py' as alias for 'boost::python' | 21 December 2017, 21:55:17 UTC |
532659d | Z Stern | 21 December 2017, 21:54:07 UTC | Comment fixes. | 21 December 2017, 21:54:07 UTC |
6ac9c6f | Steven Johnson | 21 December 2017, 21:50:11 UTC | Minor cleanups - simplify PythonListToVectorConverter to use inbuilt error handling - use 'namespace py' as abbreviation for boost::python | 21 December 2017, 21:50:11 UTC |
915420e | Z Stern | 21 December 2017, 21:45:20 UTC | Get ParamMap working with infer_input_bounds. Add test for this case. Add documentation for ParamMap functionality. Fix tabs in test. | 21 December 2017, 21:45:20 UTC |
9a1544b | Dillon Sharlet | 21 December 2017, 19:43:18 UTC | Remove temp debugging code and stale comment. | 21 December 2017, 19:43:18 UTC |
827d52d | Z Stern | 21 December 2017, 19:04:57 UTC | Merge branch 'master' into thread_safe_jit_call | 21 December 2017, 19:04:57 UTC |
529702b | Zalman Stern | 21 December 2017, 18:48:02 UTC | Merge pull request #2634 from halide/shoaibkamil/metal_struct_padding Work around Metal struct padding | 21 December 2017, 18:48:02 UTC |
d98e51b | Z Stern | 21 December 2017, 08:25:57 UTC | Fix errors passing param_map through realize layers. Add support for initializer list construction of ParamMap. Add ability to use Buffer<T> rather than just Buffer<> in setting up ParamMap. Fix test, add more test cases. | 21 December 2017, 08:25:57 UTC |
493c7b8 | dsharletg | 21 December 2017, 01:07:45 UTC | Change hexagon_benchmarks to use new convention for cross compilation. | 21 December 2017, 01:07:45 UTC |
7a64c7b | dsharletg | 21 December 2017, 00:43:49 UTC | Remove -msse2 to match Makefile. | 21 December 2017, 00:43:49 UTC |
730df45 | dsharletg | 21 December 2017, 00:43:27 UTC | Enable hexagon_matmul to run without Hexagon. | 21 December 2017, 00:43:27 UTC |
d409d8d | dsharletg | 21 December 2017, 00:21:52 UTC | Add more comments to adb_run_on_device.sh. | 21 December 2017, 00:21:52 UTC |
58495a3 | dsharletg | 21 December 2017, 00:02:26 UTC | Remove HelloHexagon app, it was almost completely redundant with the blur app. | 21 December 2017, 00:05:01 UTC |
eb043e6 | dsharletg | 20 December 2017, 23:58:33 UTC | Fix ARM intrinsics, enable running on Android. | 20 December 2017, 23:58:33 UTC |
4349324 | dsharletg | 20 December 2017, 23:02:16 UTC | Enable blur app on Android, add NEON version of blur_fast. | 20 December 2017, 23:02:16 UTC |
1ab77a7 | Dillon Sharlet | 20 December 2017, 22:51:05 UTC | Merge pull request #2655 from halide/overzealous-warning Remove spurious warning when shifting by (bits - 1). | 20 December 2017, 22:51:05 UTC |
fdad117 | Steven Johnson | 20 December 2017, 22:50:22 UTC | Merge pull request #2656 from halide/srj-lock Use ScopedSpinLock in cuda.cpp to avoid missed unlocks | 20 December 2017, 22:50:22 UTC |
2a752c9 | Z Stern | 20 December 2017, 22:22:24 UTC | Initial support for passing per-call params to realize et al. Addresses https://github.com/halide/Halide/issues/2450 | 20 December 2017, 22:22:24 UTC |
fcc8b37 | Shoaib Kamil | 20 December 2017, 21:06:39 UTC | Merge branch 'master' of https://github.com/halide/halide into shoaibkamil/metal_struct_padding | 20 December 2017, 21:06:39 UTC |
eb33292 | Shoaib Kamil | 20 December 2017, 21:06:28 UTC | Undo change, just in case it was driver wonkiness | 20 December 2017, 21:06:28 UTC |
886d5e0 | dsharletg | 20 December 2017, 20:57:12 UTC | Add Hexagon schedule. | 20 December 2017, 20:57:12 UTC |
e9b0c59 | dsharletg | 20 December 2017, 20:38:03 UTC | Remove commented/dead code. | 20 December 2017, 20:38:03 UTC |
66d9405 | Steven Johnson | 20 December 2017, 20:04:17 UTC | Use ScopedSpinLock in cuda.cpp to avoid missed unlocks | 20 December 2017, 20:04:17 UTC |
7381e4a | Dillon Sharlet | 20 December 2017, 07:38:22 UTC | Remove spurious warning when shifting by (bits - 1). | 20 December 2017, 07:38:22 UTC |
45ca86a | Steven Johnson | 20 December 2017, 01:51:37 UTC | Bring Python bindings for Halide::Var 100% up to date, with complete tests for all binding bits | 20 December 2017, 01:51:37 UTC |
8a49f4a | Steven Johnson | 20 December 2017, 01:00:58 UTC | Update Python bindings for Type Bring Python bindings for Halide::Type 100% up to date, with complete tests for all binding bits. | 20 December 2017, 01:00:58 UTC |
04db69b | Dillon Sharlet | 20 December 2017, 00:24:57 UTC | Merge pull request #2633 from halide/pdb_halide_set_par_hvx_mode Acquire qurt lock and unlock inside the pipeline code when doing parallel | 20 December 2017, 00:24:57 UTC |
dae377d | Dillon Sharlet | 20 December 2017, 00:24:29 UTC | Merge pull request #2651 from halide/target-defs2 Add #defines indicating the header is present | 20 December 2017, 00:24:29 UTC |
d4fa423 | Steven Johnson | 20 December 2017, 00:17:53 UTC | Don't call export_values() for enums | 20 December 2017, 00:17:53 UTC |
2f7f179 | Steven Johnson | 19 December 2017, 23:35:50 UTC | Remove dangly bits | 19 December 2017, 23:35:50 UTC |
b626a79 | Steven Johnson | 19 December 2017, 23:30:41 UTC | Update Python bindings for Target First attempt at bringing one of th Python bindings 100% up to date, with complete tests for all binding bits. | 19 December 2017, 23:30:41 UTC |
ff9f175 | Shoaib Kamil | 19 December 2017, 20:53:59 UTC | Merge branch 'master' of https://github.com/halide/halide into shoaibkamil/metal_struct_padding | 19 December 2017, 20:53:59 UTC |
657426b | Shoaib Kamil | 19 December 2017, 20:50:01 UTC | Check for nil function within new_compute_pipeline_state | 19 December 2017, 20:50:01 UTC |
03a7875 | Steven Johnson | 19 December 2017, 19:42:38 UTC | Merge pull request #2650 from halide/srj-tidy Fix a few minor ClangTidy warnings | 19 December 2017, 19:42:38 UTC |
34c0898 | Steven Johnson | 19 December 2017, 19:42:14 UTC | Merge pull request #2649 from halide/srj-inc Add missing #include | 19 December 2017, 19:42:14 UTC |
0f01272 | Pranav Bhandarkar | 19 December 2017, 19:39:42 UTC | Fix a bug in InjectHVXLockUnlock that was exposed by a modified version of test/correctness/parallel_nested.cpp. The bug is exposed when we have nested parallelism with alternating scopes from inner to outer that uses hvx. The modified version is now added as a new test - test/correctness/parallel_nested_1.cpp | 19 December 2017, 19:39:42 UTC |
6ef5e35 | Steven Johnson | 19 December 2017, 18:46:06 UTC | Fix a few minor ClangTidy warnings - Mismatched prototype arg names - prefer .empty() to .size() == 0 | 19 December 2017, 18:46:06 UTC |
295be42 | Steven Johnson | 19 December 2017, 18:42:06 UTC | Also fix a few clang-tidy warnings | 19 December 2017, 18:42:06 UTC |
d9e1633 | Steven Johnson | 19 December 2017, 18:26:46 UTC | Add missing #include | 19 December 2017, 18:26:46 UTC |
9c1faeb | Dillon Sharlet | 19 December 2017, 07:41:51 UTC | Merge pull request #2646 from halide/srj-opencl Reorder struct in opencl to avoid padding. | 19 December 2017, 07:41:51 UTC |
74bc5cb | Dillon Sharlet | 19 December 2017, 07:40:55 UTC | Merge pull request #2648 from halide/add--mhvx-to-simd_op_check-makefile Add -mhvx to -mhvx-double, required by our hexagon-clang++ driver | 19 December 2017, 07:40:55 UTC |
f3c9b8c | Dillon Sharlet | 19 December 2017, 05:33:39 UTC | Add #defines for when the GPU runtime is present. | 19 December 2017, 05:34:30 UTC |
01e2dfd | Ron Lieberman | 19 December 2017, 02:15:43 UTC | Add -mhvx to -mhvx-double, required by our hexagon-clang++ driver | 19 December 2017, 02:15:43 UTC |
fd1b076 | Steven Johnson | 19 December 2017, 01:43:30 UTC | Merge pull request #2641 from halide/srj-python2and3-division Update python bindings for Python2 + Python3 compatibility | 19 December 2017, 01:43:30 UTC |