c575919 | Steven Johnson | 14 February 2018, 00:05:52 UTC | "This pipeline is guaranteed to fail an assertion at runtime" should be an error, not a warning If it's guaranteed to fail, better to fail at compiletime than runtime. | 14 February 2018, 00:05:52 UTC |
31568a0 | Zalman Stern | 13 February 2018, 00:34:50 UTC | Merge pull request #2737 from halide/thread_safe_jit_call Add a test per issue https://github.com/halide/Halide/issues/2731 . | 13 February 2018, 00:34:50 UTC |
483e70d | Z Stern | 12 February 2018, 21:03:37 UTC | Add a comment to indicate why a mutex is acquired in the case where only one thread is compiling. | 12 February 2018, 21:03:37 UTC |
923d651 | Z Stern | 11 February 2018, 20:02:14 UTC | Add a test per issue https://github.com/halide/Halide/issues/2731 . Also serves as documnetation on passing parameters to JITted routines and avoiding recompilation for each realize. Suggestions on further comments for documentation are particularly welcome. | 11 February 2018, 20:02:14 UTC |
5125fa3 | Andrew Adams | 09 February 2018, 23:05:04 UTC | Merge pull request #2733 from halide/ger_cleanup reschedule ger | 09 February 2018, 23:05:04 UTC |
b7176fb | Andrew Adams | 09 February 2018, 17:09:28 UTC | parallel -> true | 09 February 2018, 17:09:28 UTC |
1a83604 | Andrew Adams | 08 February 2018, 22:13:46 UTC | reschedule ger Change ger to be in-place, and simplify the schedule. Performance now closer to openblas for me on the larger sizes, and is much faster on the smaller sizes. Rightmost column is proportional to flops Halide sger 32 0.081833 13.2953 Halide sger 64 0.226842 18.6209 Halide sger 128 1.073608 15.4991 Halide sger 288 7.498786 11.1378 Halide sger 544 29.680935 10.0072 Halide sger 1056 121.922739 9.16357 Halide sger 2080 1147.310148 3.77453 OpenBLAS sger 32 35.603869 0.0305585 OpenBLAS sger 64 68.888618 0.0613164 OpenBLAS sger 128 43.106477 0.386021 OpenBLAS sger 288 92.307125 0.904806 OpenBLAS sger 544 186.297631 1.59435 OpenBLAS sger 1056 363.545260 3.0732 OpenBLAS sger 2080 1035.606023 4.18167 | 08 February 2018, 22:13:46 UTC |
bef8bc2 | Zalman Stern | 08 February 2018, 17:30:30 UTC | Merge pull request #2730 from adityaatluri/rocm-make added makefile changes to build for amdgpu | 08 February 2018, 17:30:30 UTC |
aad17fb | Steven Johnson | 08 February 2018, 00:38:52 UTC | Merge pull request #2719 from halide/srj-pycallgen Augment the Python Bindings to allow using precompiled Generators | 08 February 2018, 00:38:52 UTC |
e604193 | Steven Johnson | 07 February 2018, 21:54:37 UTC | Merge branch 'master' into srj-pycallgen | 07 February 2018, 21:54:37 UTC |
60c5e5c | Steven Johnson | 07 February 2018, 21:52:31 UTC | Merge branch 'master' into srj-pycallgen | 07 February 2018, 21:52:31 UTC |
ea1d830 | Steven Johnson | 07 February 2018, 21:52:04 UTC | Merge pull request #2728 from halide/srj-llset Revise GeneratorParam<LoopLevel>::set() work for both before-and-after generate() usage | 07 February 2018, 21:52:04 UTC |
bbada39 | Steven Johnson | 07 February 2018, 18:14:10 UTC | Merge branch 'master' into srj-llset | 07 February 2018, 18:14:10 UTC |
f8c177b | Steven Johnson | 07 February 2018, 18:13:40 UTC | Merge pull request #2729 from halide/srj-unused Remove unused 'using's | 07 February 2018, 18:13:40 UTC |
6b3df35 | Aditya Atluri | 07 February 2018, 03:09:26 UTC | changed cmake in src directory to enable WITH_AMDGPU macro for compilation | 07 February 2018, 03:09:26 UTC |
10a9b3e | Aditya Atluri | 07 February 2018, 02:59:24 UTC | Made changes to cmake to build amdgpu and added comments about amdgpu build status | 07 February 2018, 02:59:24 UTC |
f31df5f | Aditya Atluri | 07 February 2018, 02:41:37 UTC | added makefile changes to build for amdgpu 1. Changed Makefile to enable compile flag WITH_AMDGPU 2. Changed CodeGen_LLVM file to check if intializing amdgpu works or not | 07 February 2018, 02:41:37 UTC |
504a7fc | Steven Johnson | 06 February 2018, 23:57:55 UTC | Remove unused 'using's | 06 February 2018, 23:57:55 UTC |
7c5d674 | Steven Johnson | 06 February 2018, 23:51:11 UTC | Add explicit Func name | 06 February 2018, 23:51:11 UTC |
a8a2178 | Steven Johnson | 06 February 2018, 23:50:48 UTC | Add explicit Func name | 06 February 2018, 23:50:48 UTC |
b4bdb05 | Steven Johnson | 06 February 2018, 23:17:12 UTC | Patch PR#2728 | 06 February 2018, 23:17:12 UTC |
2f4c511 | Steven Johnson | 06 February 2018, 23:16:42 UTC | Add missing 'using' | 06 February 2018, 23:16:42 UTC |
95a4c88 | Steven Johnson | 06 February 2018, 23:05:03 UTC | Merge pull request #2727 from halide/srj-tupleselect tuple_select cleanup | 06 February 2018, 23:05:03 UTC |
979b56e | Steven Johnson | 06 February 2018, 23:01:45 UTC | Merge branch 'master' into srj-pycallgen | 06 February 2018, 23:01:45 UTC |
e207f95 | Steven Johnson | 06 February 2018, 23:00:39 UTC | Revise GeneratorParam<LoopLevel>::set() work for both before-and-after generate() usage A much-simpler alternative to PR#2720. Note that the interesting change is really all in the override to set() in GeneratorParam_LoopLevel (the changes to the base class implementation are just drive-by simplification to make that override a bit easier to reason about). | 06 February 2018, 23:00:39 UTC |
0e001cc | Steven Johnson | 06 February 2018, 20:01:21 UTC | Clarify and rename methods related to Outputs | 06 February 2018, 20:01:21 UTC |
2ba8955 | Steven Johnson | 06 February 2018, 19:23:48 UTC | Remove unused stuff | 06 February 2018, 19:23:48 UTC |
93ea595 | Steven Johnson | 06 February 2018, 19:18:56 UTC | Fix comment | 06 February 2018, 19:18:56 UTC |
74c4398 | Steven Johnson | 06 February 2018, 18:26:28 UTC | Restore all-are-convertible test | 06 February 2018, 18:26:28 UTC |
f1b7707 | Steven Johnson | 06 February 2018, 18:13:32 UTC | Merge branch 'master' into srj-tupleselect | 06 February 2018, 18:13:32 UTC |
18cd57e | Steven Johnson | 05 February 2018, 21:59:11 UTC | Move non-templated select() and tuple_select() impls into IROperator.cpp | 05 February 2018, 21:59:11 UTC |
137e2b2 | Zalman Stern | 05 February 2018, 21:51:20 UTC | Merge pull request #2726 from halide/zero_init_buffer_shape zero-initialize all fields of a Halide::Runtime::Buffer | 05 February 2018, 21:51:20 UTC |
9709513 | Steven Johnson | 05 February 2018, 21:03:26 UTC | tuple_select cleanup -- move to IROperator.h, next to select(), for better discoverability -- add multiway verison of tuple_select(), to match select() -- add test | 05 February 2018, 21:03:26 UTC |
b7959eb | Andrew Adams | 05 February 2018, 19:14:35 UTC | Try to avoid gcc 4.8 internal compiler error | 05 February 2018, 19:14:35 UTC |
ef8ff74 | Andrew Adams | 05 February 2018, 17:50:48 UTC | zero-initialize all fields of a Halide::Runtime::Buffer | 05 February 2018, 17:50:48 UTC |
8cdc8cd | Andrew Adams | 05 February 2018, 02:29:13 UTC | Merge pull request #2725 from halide/export_codegen_internal Don't include LLVM_Headers in CodeGen_Internal.h | 05 February 2018, 02:29:13 UTC |
41157da | Andrew Adams | 04 February 2018, 03:07:00 UTC | Don't include LLVM_Headers in CodeGen_Internal.h, so that it can be put in Halide.h | 04 February 2018, 03:07:00 UTC |
db4927a | Andrew Adams | 04 February 2018, 00:32:11 UTC | Merge pull request #2724 from halide/simplify_const_cast Remove unnecessary template function | 04 February 2018, 00:32:11 UTC |
980daf9 | Andrew Adams | 03 February 2018, 02:50:51 UTC | Remove unnecessary template function | 03 February 2018, 02:50:51 UTC |
81de819 | Steven Johnson | 03 February 2018, 01:50:05 UTC | Merge pull request #2722 from halide/srj-cast Fix error in commit c2fd18 | 03 February 2018, 01:50:05 UTC |
d414e1b | Steven Johnson | 02 February 2018, 19:05:06 UTC | Fix error in commit c2fd18 copy-paste error; several of the changes should have been <void> rather than <T>, since remove_const must match the type. | 02 February 2018, 19:05:06 UTC |
5e432c6 | Steven Johnson | 02 February 2018, 02:06:47 UTC | Merge pull request #2721 from halide/srj-tidy Fix minor clang-tidy warnings | 02 February 2018, 02:06:47 UTC |
71f8594 | Steven Johnson | 01 February 2018, 23:14:16 UTC | Fix minor clang-tidy warnings Various declaration-vs-definition-names-differ warnings to silence downstream nags | 01 February 2018, 23:14:16 UTC |
6c00f11 | Steven Johnson | 31 January 2018, 23:33:31 UTC | Turn off debugging | 31 January 2018, 23:33:31 UTC |
9518686 | Steven Johnson | 30 January 2018, 21:50:11 UTC | Augment the Python Bindings to allow using a precompiled Generator (written in C++) at JIT time via the Python bindings, mimicking the C++ stub functionality. | 31 January 2018, 23:30:49 UTC |
6908edb | Steven Johnson | 31 January 2018, 17:27:09 UTC | Merge pull request #2717 from halide/srj-movereg Add move ctor/op= for Parameter | 31 January 2018, 17:27:09 UTC |
3b0e1dc | Andrew Adams | 31 January 2018, 16:09:48 UTC | Merge pull request #2715 from halide/add_rpath_for_llvm When building Halide without llvm, add an rpath to the llvm used | 31 January 2018, 16:09:48 UTC |
528ec2b | Steven Johnson | 31 January 2018, 01:24:36 UTC | Add move ctor/op= for Parameter The implicitly-define move ctor for Parameter didn't handle object instance registry tracking, which is must do. (Other usages should never be copyable or movable; move ops were deleted to ensure this.) | 31 January 2018, 01:24:36 UTC |
658e965 | Andrew Adams | 30 January 2018, 22:16:33 UTC | Add explanatory comment | 30 January 2018, 22:16:33 UTC |
22a85d1 | Zalman Stern | 30 January 2018, 21:22:16 UTC | Merge pull request #2716 from halide/srj-ret in_gpu_loop wasn't being restore | 30 January 2018, 21:22:16 UTC |
56a5608 | Steven Johnson | 30 January 2018, 18:08:29 UTC | in_gpu_loop wasn't being restore See previous line 454. Switched all usage to ScopedValue to avoid needing manual cleanup. | 30 January 2018, 18:08:29 UTC |
edab342 | Jing Pu | 30 January 2018, 04:06:59 UTC | Merge pull request #2714 from halide/lift_reinterpret Allow substituting reinterpret calls back in LICM. | 30 January 2018, 04:06:59 UTC |
17bad9b | Andrew Adams | 30 January 2018, 02:28:27 UTC | When building Halide without llvm, add an rpath to the llvm used | 30 January 2018, 02:28:27 UTC |
68e2538 | Andrew Adams | 30 January 2018, 01:20:06 UTC | Merge pull request #2645 from halide/export_all Expose internal symbols | 30 January 2018, 01:20:06 UTC |
6d26b97 | Andrew Adams | 29 January 2018, 22:18:07 UTC | A single space character counts as true, apparently | 29 January 2018, 22:18:07 UTC |
022f01a | Jing Pu | 29 January 2018, 21:59:18 UTC | Allow lifting reinterpret calls in LICM. | 29 January 2018, 21:59:18 UTC |
92cedc3 | Andrew Adams | 29 January 2018, 19:46:56 UTC | Take RTTI settings from LLVM by default But allow the configuration LLVM with RTTI + Halide without RTTI, via WITH_RTTI in the Makefile and HALIDE_ENABLE_RTTI in cmake | 29 January 2018, 19:46:56 UTC |
9730393 | Andrew Adams | 27 January 2018, 04:11:16 UTC | export symbols on linux with cmake | 27 January 2018, 04:11:16 UTC |
bc83c0f | Andrew Adams | 26 January 2018, 22:45:47 UTC | Missing endif | 26 January 2018, 22:45:47 UTC |
90c092a | Andrew Adams | 26 January 2018, 22:36:07 UTC | Adding dllexport to global vars | 26 January 2018, 22:36:07 UTC |
16328f0 | Andrew Adams | 26 January 2018, 21:23:02 UTC | Merge remote-tracking branch 'origin/master' into export_all | 26 January 2018, 21:23:02 UTC |
9b03ec7 | Andrew Adams | 26 January 2018, 21:20:01 UTC | Flip sense of incorrect ifdef | 26 January 2018, 21:20:01 UTC |
e40ffc8 | Andrew Adams | 26 January 2018, 21:18:02 UTC | Merge pull request #2526 from halide/cuda_register_shuffle Support for warp shuffle instructions on cuda | 26 January 2018, 21:18:02 UTC |
a3a318f | Andrew Adams | 25 January 2018, 22:10:06 UTC | Forgot to actually define HALIDE_EXPORT | 25 January 2018, 22:10:06 UTC |
68a2c60 | Andrew Adams | 25 January 2018, 18:28:08 UTC | Spacing fix | 25 January 2018, 18:28:08 UTC |
9a08969 | Andrew Adams | 25 January 2018, 18:26:15 UTC | Indentation fix | 25 January 2018, 18:26:15 UTC |
b6595bd | Andrew Adams | 25 January 2018, 18:10:32 UTC | Remove EXPORT on functions only | 25 January 2018, 18:10:32 UTC |
59bca3c | Andrew Adams | 25 January 2018, 16:38:19 UTC | Review comments | 25 January 2018, 16:38:19 UTC |
ce02cd9 | Andrew Adams | 25 January 2018, 16:31:02 UTC | Merge remote-tracking branch 'origin' into export_all | 25 January 2018, 16:31:02 UTC |
eeaae61 | Andrew Adams | 25 January 2018, 16:30:05 UTC | Merge pull request #2695 from halide/clone_shares_cache_key A clone of a Func should share cache entries with the original | 25 January 2018, 16:30:05 UTC |
446267d | Andrew Adams | 25 January 2018, 16:29:47 UTC | Merge pull request #2710 from halide/srj-ub Avoid UBSan failures | 25 January 2018, 16:29:47 UTC |
ac5ca9d | Steven Johnson | 25 January 2018, 03:46:25 UTC | Avoid UBSan failures Initing a Buffer<> field to an undef Buffer<> value (e.g. from calling Variable::make) triggers UBSan failures (referencing to an offset of a null contents ptr). | 25 January 2018, 03:46:25 UTC |
e3d1023 | Andrew Adams | 25 January 2018, 01:25:27 UTC | Merge branch 'master' into clone_shares_cache_key | 25 January 2018, 01:25:27 UTC |
4ddd173 | Andrew Adams | 25 January 2018, 01:24:25 UTC | Merge pull request #2639 from halide/nested_tailstrategy_auto_bug Fix unwanted TailStrategy::Auto behavior in corner case | 25 January 2018, 01:24:25 UTC |
7307307 | Andrew Adams | 24 January 2018, 22:45:30 UTC | Try exporting all on every platform | 24 January 2018, 22:45:30 UTC |
5ae237a | Andrew Adams | 24 January 2018, 22:42:15 UTC | Merge branch 'master' into export_all | 24 January 2018, 22:42:15 UTC |
262ee9e | Andrew Adams | 24 January 2018, 22:41:24 UTC | Merge branch 'master' into cuda_register_shuffle | 24 January 2018, 22:41:24 UTC |
c36879c | Andrew Adams | 24 January 2018, 22:39:07 UTC | Address review comments | 24 January 2018, 22:39:07 UTC |
8d428e0 | Steven Johnson | 24 January 2018, 19:58:44 UTC | Merge pull request #2707 from halide/srj-pil Add ImageParam, OutputImageParam to py bindings | 24 January 2018, 19:58:44 UTC |
d20118c | Steven Johnson | 24 January 2018, 17:16:28 UTC | Remove the is_const enforcement from Dimension | 24 January 2018, 17:16:28 UTC |
b7337a4 | Steven Johnson | 24 January 2018, 17:11:00 UTC | Merge branch 'master' into srj-pil | 24 January 2018, 17:11:00 UTC |
02972de | Zalman Stern | 23 January 2018, 21:22:58 UTC | Merge pull request #2706 from halide/srj-pybind11 Remove python_bindings/doc | 23 January 2018, 21:22:58 UTC |
d469718 | Steven Johnson | 23 January 2018, 20:08:53 UTC | Merge pull request #2709 from halide/srj-argmax Fix bonehead copy-paste mistake in Python wrappers | 23 January 2018, 20:08:53 UTC |
d12741f | Steven Johnson | 23 January 2018, 06:24:43 UTC | Fix bonehead copy-paste mistake in Python wrappers | 23 January 2018, 06:24:43 UTC |
805657d | Steven Johnson | 19 January 2018, 22:29:13 UTC | Merge pull request #2696 from halide/srj-stderr posix_io outputs to stdout instead of stderr (Issue #2686) | 19 January 2018, 22:29:13 UTC |
93705fd | Steven Johnson | 19 January 2018, 18:32:11 UTC | Merge branch 'master' into srj-pil | 19 January 2018, 18:32:11 UTC |
1635251 | Steven Johnson | 19 January 2018, 17:57:43 UTC | make_query -> make_bounds_query | 19 January 2018, 17:57:43 UTC |
d15d422 | Steven Johnson | 19 January 2018, 02:01:58 UTC | Merge pull request #2705 from halide/srj-depr Remove long-deprecated gpu_tile() methods, Take 2 | 19 January 2018, 02:01:58 UTC |
07be31b | Steven Johnson | 19 January 2018, 01:09:19 UTC | Add ImageParam, OutputImageParam to py bindings Added OutputImageParam and Dimension and finished ImageParam; added limited (but incomplete) testing. Note the change to Dimension in the core code: the issue here is that in an attempt to prevent a const-Dimension from being copied into a non-const version and later mutated, we made the copy ctor private with friend access. This is untenable for the Python bindings; machinery inside PyBind11 needs to be able to copy/move them, and declaring 'friend' access would be a nightmare. Instead, make it public, and add the constness checks via a runtime flag. | 19 January 2018, 01:09:19 UTC |
fba1d15 | Steven Johnson | 18 January 2018, 23:22:05 UTC | Remove python_bindings/doc Hopelessly out of date, will be revisited at a later date. Also removed cruft from requirements.txt | 18 January 2018, 23:22:05 UTC |
b029df2 | Steven Johnson | 18 January 2018, 22:45:37 UTC | Merge pull request #2704 from halide/srj-pybind11 Fix pybind's buffer-info lifetime management | 18 January 2018, 22:45:37 UTC |
60e8cbb | Steven Johnson | 18 January 2018, 22:45:23 UTC | Updated comment. | 18 January 2018, 22:45:23 UTC |
850bf95 | Steven Johnson | 18 January 2018, 22:44:18 UTC | Remove long-deprecated gpu_tile() methods, Take 2 Removing them once again, and also updating the other variants that were split into multiple versions as a result. | 18 January 2018, 22:44:18 UTC |
e0778ce | Steven Johnson | 18 January 2018, 19:59:03 UTC | Fix pybind's buffer-info lifetime management The previous pybind code didn't properly manage the lifetime of Python buffer objects when used to construct a Buffer in a shared way. We now use a PyBind alias object for buffers to ensure the lifetime is managed properly. | 18 January 2018, 19:59:03 UTC |
43f8ad6 | Steven Johnson | 18 January 2018, 18:46:42 UTC | Merge pull request #2702 from halide/srj-copyconst make_with_shape_of() didn't handle const types correctly | 18 January 2018, 18:46:42 UTC |
c2fd186 | Steven Johnson | 18 January 2018, 07:16:08 UTC | Add extra code to avoid "casting away const" on some compilers | 18 January 2018, 07:16:08 UTC |
25d048f | Steven Johnson | 18 January 2018, 01:45:04 UTC | make_with_shape_of() didn't handle const types correctly Injected by https://github.com/halide/Halide/pull/2677 | 18 January 2018, 01:45:04 UTC |
71dd08a | Steven Johnson | 17 January 2018, 23:12:50 UTC | Merge pull request #2700 from halide/revert-2697-srj-depr Revert "Remove long-deprecated variants of gpu_tile()" | 17 January 2018, 23:12:50 UTC |
1b3b995 | Steven Johnson | 17 January 2018, 23:10:42 UTC | Revert "Remove long-deprecated variants of gpu_tile()" | 17 January 2018, 23:10:42 UTC |
b5c63a5 | Steven Johnson | 17 January 2018, 18:41:20 UTC | Merge pull request #2688 from halide/srj-pybind11 Rewrite python_bindings to use PyBind11 instead of Boost.Python | 17 January 2018, 18:41:20 UTC |