swh:1:snp:70f530b74f5be73cfb71c212c9e3317ce44c1ebc

sort by:
Revision Author Date Message Commit Date
346c37b Reset branch onto new history Was ab17f01fc00f31db5ed5358307bf39501cb698b4 07 December 2017, 21:54:28 UTC
8a00cac Minor fix to vectorization Former-commit-id: 0a8b31b221dcdc184304e428f73da14ea3cee8ad 16 December 2016, 00:00:24 UTC
78fb337 Fix passing bool types into gpu kernels Former-commit-id: 831bd12bc7a9bb77951e52c6365b4f66f12aaa6c 15 December 2016, 22:46:11 UTC
9a39e0f Merge pull request #1689 from halide/no_autovectorize_in_runtime Disable early autovectorization of runtime Former-commit-id: bf81d73b2629646bf91392b2ccda1a0683c413b6 15 December 2016, 21:50:41 UTC
90356ab Merge pull request #1692 from halide/float_api_fix Adapt to new Float api for llvm_ver >= 40 Former-commit-id: e7faa786c6ce050bec0ef24d671f3eba7f32f34d 15 December 2016, 18:31:27 UTC
6796522 Adapt to new Float api for llvm_ver >= 40 Former-commit-id: 97d4e993ec4393f47e5dbb695ac4ac29e34e112c 15 December 2016, 15:07:19 UTC
0508e86 Merge pull request #1657 from halide/hexagon-power Added two more methods for powering on HVX Former-commit-id: 065e872fea7d42faef79171c4b8614d757831f20 14 December 2016, 22:45:34 UTC
ee327f9 Merge pull request #1682 from halide/hex-add-sim-readme Add addtional needed path for hexagon LD_LIBRARY_PATH Former-commit-id: 1e838002086368a5ef371a1b9135d4ea4e472b67 14 December 2016, 19:34:31 UTC
087b9c0 use relative path Former-commit-id: 40ab1048ce54695fef7549adfe616babf2c964aa 14 December 2016, 19:29:08 UTC
cdfedf5 Merge pull request #1686 from halide/HelloHexagon_Buffer Update HelloHexagon to use Halide::Buffer Former-commit-id: dfff62e09b352aa71b8b0d8c15a5806d691ba313 14 December 2016, 02:13:44 UTC
d126703 DRY Former-commit-id: bea16cd79ba7794bebb8f15ebae08e056c45dcd4 14 December 2016, 02:10:30 UTC
8b2aaa6 Disable early autovectorization of runtime We compile the runtime modules to .ll for a "generic" target at -O3. Apparently in some circumstances this autovectorizes to 128-bits. Not all vector types are supported by all targets (hexagon), so this is bad, and can cause assertion failures inside LLVM during instruction selection. This PR turns this off. The .ll gets a second chance to autovectorize later in codegen once the module is assembled and has a concrete target. Former-commit-id: 457ee7eaa6246419a2a2d0e5cd55f0a0279dd081 14 December 2016, 02:08:24 UTC
6157aa0 Merge pull request #1679 from halide/fix_openglcompute_app Fix openglcompute app Former-commit-id: e9d4b85b91084bc05fb30012b6672c33a1d882de 13 December 2016, 22:43:36 UTC
816fa8b Update HelloHexagon to use Halide::Buffer Former-commit-id: 10a4b5f2920c9f64bf4f40bc21c2caebf6fe308a 13 December 2016, 22:41:52 UTC
9d1ba1d Merge pull request #1674 from halide/oip Remove deprecated methods from OutputImageParam, plus usage fixes Former-commit-id: 8978a40d9736ebe41ba5f959736bc200c2dbaff9 13 December 2016, 22:09:02 UTC
6528763 Merge pull request #1665 from halide/BufferRefCollection Fix realizing into multiple existing buffers Former-commit-id: 028f1609fddae010d3cbd5bf095aac2127a372f4 13 December 2016, 22:08:17 UTC
2dae28c Remove pointless overloads Former-commit-id: 86354b489221e896eea774c31c3112d042d87689 13 December 2016, 22:07:54 UTC
2d3eec5 Merge pull request #1678 from halide/dim Move Dimension from OutputImageParam into Parameter.h Former-commit-id: 2aeb9694610e9ef2f34355919bf0d173d53aa45d 13 December 2016, 20:54:56 UTC
7a2103b Vectorized pred test should only run on X86 or Hexagon Former-commit-id: c5b8f369e0be000e61e6301a4af491e6e73baab5 13 December 2016, 20:15:14 UTC
d3bef21 Merge pull request #1684 from halide/halide-buffer-translated-fix Set correct return type for Buffer::translated Former-commit-id: 98b909f8f44d64a23cfe7f1a988bb4772cddd8f2 13 December 2016, 19:51:07 UTC
f1d7515 Set correct return type for Buffer::translated Astonishingly, this was caught by nvcc, of all compilers. /attn @abadams Former-commit-id: 6b5146569525da3869dbe0fdeece995c04936e47 13 December 2016, 19:47:46 UTC
93151be Merge pull request #1683 from halide/lesson10 lesson 10 was broken for CMake build Former-commit-id: 5452aca8d117c9782e6ce0eef381e0952b671a5b 13 December 2016, 19:38:31 UTC
1410445 lesson 10 was broken for CMake build Former-commit-id: 70fbddcfef5de3356bebc06a61a78168ec9bfec1 13 December 2016, 18:05:43 UTC
d38fa4f document need for HL_HEXAGON_SIM_REMOTE in README.md Former-commit-id: cdedf1a3a8049228bf16016f14bf572bb438d897 13 December 2016, 16:41:35 UTC
89a28e3 Add addtional needed path for hexagon LD_LIBRARY_PATH Former-commit-id: 0accbe887955c21882f40673bdbd9d200dd0c347 13 December 2016, 15:31:13 UTC
ed9b64c Merge pull request #1345 from halide/vectorized_pred Vectorized pred Former-commit-id: bc334ee1bf2384e7017feb8ef3166ca22c759b1f 13 December 2016, 01:58:33 UTC
98e6068 Merge pull request #1680 from halide/glsl_app_buffers Update opengl_test to use Halide::Buffer Former-commit-id: f8dcd439722cd862222a9f504b343de63d9218d4 13 December 2016, 01:19:50 UTC
54921a2 Update opengl_test to use Buffer Former-commit-id: e6c3811cd8478c1d0d541746f849e9ba8e6a004a 13 December 2016, 00:50:41 UTC
12b1305 Fix openglcompute app Made it use Halide::Buffer. While doing this I discovered that the vector support in OpenGLCompute codegen was broken, and made a dangerous assumption anyway (all vector access is aligned), so I neutered it for now. It needs a pass that actually detects if all buffer access is aligned dense vectors, and if so, declares the buffer as an array of vectors. Former-commit-id: 475808a7d314d21d0d74da7ea7b1333627573fb8 13 December 2016, 00:46:38 UTC
bd7b692 Move Dimension from OutputImageParam into Parameter.h I need to re-use it for Generator purposes; mostly-decoupling it from OutputImageParam makes that less weird. (“Mostly” because we still grant friend access for ctors, though frankly, we could probably just make them public) Former-commit-id: 3c7c445978ae206f291f669875c71bdf01659c94 13 December 2016, 00:03:24 UTC
401ff52 Moved comments Former-commit-id: ff102acf4ce6b560481c840deef40bc1a927cc1c 12 December 2016, 23:05:28 UTC
ad19e53 Merge branch 'master' of https://github.com/halide/Halide into hexagon-power Former-commit-id: acb4110fc9a6af348bd7a8866e816ca55ad529bb 12 December 2016, 22:26:31 UTC
bc1aebf Merge branch 'master' of ssh://github.com/halide/Halide Former-commit-id: 9e9e94cd8472455805040f28ae21371e0660c521 12 December 2016, 22:24:49 UTC
b036704 Fix compilation of simd_op_check Former-commit-id: bea39bd064b4e9735b6627c3cef77facc4266346 12 December 2016, 22:24:43 UTC
8568fd4 Add memset to fake dlsym implementation. Former-commit-id: b0837c4983d4c749dccc9eff2397db063f82e3f5 12 December 2016, 22:09:35 UTC
6f1c449 Merge branch 'master' of github.com:halide/Halide into vectorized_pred Former-commit-id: c67f93018ff713b44075f5aba21a54c2c870a9a6 12 December 2016, 22:00:42 UTC
f908350 Clean up code and stale comments Former-commit-id: 7b22db865ed8192711ce3c418bf75ba3ac017244 12 December 2016, 21:14:50 UTC
1a1d12c Merge pull request #1673 from halide/remove_buffer_t_from_tutorials Remove buffer_t from tutorials Former-commit-id: 003b48383b378486e5dd5130d3941f7cc2ba0874 12 December 2016, 21:08:10 UTC
1fe17d9 Merge pull request #1677 from halide/ptx cxx_mangling_gpu should be if(TARGET_PTX) (Issue #1676) Former-commit-id: 23a9a613bee92a86ed1054c66c3a550cb3bff32a 12 December 2016, 18:22:26 UTC
0c59c15 cxx_mangling_gpu should be if(TARGET_PTX) Former-commit-id: 48daf610120cbf6a121e13d45b9f07e36fa2febf 12 December 2016, 18:14:22 UTC
9076c54 Merge pull request #1675 from halide/fix_ptx_passes Fix PTX pass manager. Former-commit-id: 5ddf3ba287d1593749c9a626ef1d84817251f7a2 12 December 2016, 18:12:15 UTC
75fd1b4 Fine tune performance parameters - increase low max_bus_bw while preserving bits - decrease bus usage percentage for low & nominal - increase bus bw for turbo (safety margin) - sink mipsTotal computation outside of switch Former-commit-id: d2dcca5311b009d3ad349ae02ff9642b4078c458 12 December 2016, 17:52:13 UTC
e894122 Another slight tweak Former-commit-id: e403404d39a8f7c7197eb657fcc276aaf330a8b5 11 December 2016, 01:57:45 UTC
b1c9bad Slight schedule improvement Former-commit-id: 57a6d34ff3c8cbe98303a8c79053c049cdd11031 11 December 2016, 01:53:48 UTC
ea8e303 Slightly better schedule Former-commit-id: a7cfdbde2b68d8c930f0a7f1ec9c7726ccdda5ca 10 December 2016, 18:11:26 UTC
b55c82d Fix PTX pass manager. LLVM was perfectly capable of fragmenting allocations if we just asked it to. Former-commit-id: 0e1662f2382e7134205abcdcd995a54f3441365a 10 December 2016, 04:01:04 UTC
abedbcd Update from 12/8 review comments: halide_hexagon_remote_power_hvx_on_perf(): - change bwBytePerSec to bwMegabytesPerSec - remove unused result halide_hexagon_remote_power_hvx_on_mode(): - don't check power_ref_count when computing values - hoist values out of switch that are set the same in all modes - verified turbo performance stil matches tip (camera_pipe & HelloHexagon) - verified low, nominal, & turbo power modes scale well - camera_pipe increases in performance ~20% on each step Former-commit-id: e093d22a78460718dda221660455da8b7d1444f8 10 December 2016, 03:31:03 UTC
8e1a2ec Revise HALIDE_ATTRIBUTE_DEPRECATED to use message Former-commit-id: 62693b854e3f4e548ed4489bc8c5854823eb09ca 09 December 2016, 23:09:21 UTC
9bb2b35 Missed two Former-commit-id: 6d9b815ef085705bc6c4be0c939247f350c4f425 09 December 2016, 23:05:17 UTC
1808dfd define HALIDE_ATTRIBUTE_DEPRECATED; mark OutputImageParam methods with it (rather than deleting them) Former-commit-id: 6d87eddf98ab9cabfa69fc86635e74a60984cd04 09 December 2016, 22:43:15 UTC
0aa7719 Rename BufferRefCollection to BufferRefs Former-commit-id: 80348692038ac335642c215e176b8eb6b0337c94 09 December 2016, 22:39:20 UTC
0cadff4 Merge branch 'master' into BufferRefCollection # Conflicts: # src/Pipeline.cpp Former-commit-id: 4806c08f83797bdc79c190751cef2b66075f6462 09 December 2016, 22:36:47 UTC
c31178e Merge branch 'master' into oip # Conflicts: # tutorial/lesson_16_rgb_generate.cpp Former-commit-id: e98d273ac8fcd35e3e2c3563eab4b18fb1df9f05 09 December 2016, 22:26:26 UTC
55c0684 Remove deprecated methods from OutputImageParam, plus usage fixes Former-commit-id: f65aedc091ad4d5222ef1df610482051d1f44f77 09 December 2016, 22:25:31 UTC
0969337 Remove buffer_t from tutorials We now want people to always just use Halide::Buffer. Encouraging people to use naked buffer_t was a mistake - everyone just wrote their own wrapper classes. Former-commit-id: 84bda3d1748e826d6f43d66aa7d9d3f68f4b47fb 09 December 2016, 22:21:13 UTC
4a1ddcd Merge pull request #1670 from halide/gpp fix C++ name-mangling for GPU targets Former-commit-id: 3be9cf91db0181efa7f5422cab8347524f365585 09 December 2016, 22:01:12 UTC
010fa69 Fix tutorial convention Former-commit-id: 15f72946b92ba9c71072dad2fb7f63a89d98a992 09 December 2016, 22:00:02 UTC
58df70f Upgrade some tests to Buffer<> Former-commit-id: 58d361499c8c5c180ebc9d9356fe8e422cb7da15 09 December 2016, 21:59:57 UTC
1e75589 Merge branch 'master' of ssh://github.com/halide/Halide Former-commit-id: 3df83169eee01483ce0e587c5ab7a056664e9293 09 December 2016, 21:30:44 UTC
46aacdc Remove some explicit buffer use from linear algebra Former-commit-id: 3111e0bc01368d20983212c04e200913cbdf8ce2 09 December 2016, 21:30:38 UTC
4a7cba0 Merge pull request #1672 from halide/bufde Buffer::decref()/decref_dev() need to null alloc and buf.host Former-commit-id: 7bc335e9789615aba25b07f1e9c0adcf5c461a38 09 December 2016, 21:23:38 UTC
f925bdb Unconditionally zero alloc in decref and decref_dev Former-commit-id: 65543184be3d03d216e5f6200524ab0562dcf128 09 December 2016, 21:13:12 UTC
82f5553 Also null dev_ref_count in device_deallocate() Former-commit-id: 8bbad2d6ba36b0641c2df095c20983334b91e2dc 09 December 2016, 20:25:24 UTC
1e9288f Buffer::deallocate() needs to null alloc and buf.host Otherwise explicitly calling deallocate() could lead to double-frees later. (Note: do we need a similar fix in device_deallocate()?) Former-commit-id: 459e6c36d558de2115c9a11aa7e81bd0798fe6d2 09 December 2016, 20:24:06 UTC
d4da791 Type fix. Former-commit-id: 86690217a032b336521636c436fbb7db050c92cb 09 December 2016, 19:47:22 UTC
feb0f9c Another type fix Former-commit-id: 6c3673ee127e13714a8df999795a446927630943 09 December 2016, 19:41:47 UTC
99256ad fix type on halide_error_buffer_allocation_too_large Former-commit-id: fbe3ce78e228c9c24dae374d2952ee2ed292d405 09 December 2016, 19:31:11 UTC
cfc0be4 Remove pointless extern C Former-commit-id: 4767439f1107ef7cd837ac000995877490116b4f 09 December 2016, 19:09:25 UTC
7231cfc Fix types in tiled_blur test Former-commit-id: 3e4e5821b69c89b97faaf287105e36851289e7f4 09 December 2016, 18:42:28 UTC
d86d3e1 Remove dead code Former-commit-id: cdd1c4a060c9c14ab650905369194f8ff914b556 09 December 2016, 18:42:12 UTC
7b5dd09 Add missing _t to halide_device_interface Former-commit-id: 159c11e83de9f9adca48011cefbb57d205392f63 09 December 2016, 18:21:29 UTC
c700eb5 Revamp tests to only require linkage, not correct runtime execution Former-commit-id: ca566e219fdaa5c274e0801f48afe517c933d5b4 09 December 2016, 00:53:49 UTC
9d267f1 Merge branch 'master' into gpp Former-commit-id: 1ef21ef3e944f5e87de70585bca74af5e71e89dd 09 December 2016, 00:07:33 UTC
bbc6c0f Merge pull request #1664 from halide/require Add require() operator Former-commit-id: 8511535089b45f60425532955824edaa46d03d54 08 December 2016, 23:38:14 UTC
102b138 Merge pull request #1667 from halide/cuda_mat_mul_tests cuda matrix multiply fixes Former-commit-id: 1d9450748cb0236c996a0976ae7f0dc0beac870d 08 December 2016, 23:30:45 UTC
11f0da8 Fix typo Former-commit-id: b3448dc59d9ef1941d43807a3a3578bebc7df87c 08 December 2016, 22:27:41 UTC
812880f fix C++ name-mangling for GPU targets We never name-mangled the external functions if the target had a GPU feature; we never noticed this because we didn’t attempt to test it. Former-commit-id: f67e691870b57b3ccae3334e08df9397cf766f53 08 December 2016, 22:23:20 UTC
402a453 Update from review comments: halide_hexagon_remote_power_hvx_on_perf(): - replace reinterpret casts, with unsigned long [long] in IDL - inline halide_hexagon_HAP_power_hvx_on - only increment power_ref_count in halide_hexagon_remote_power_hvx_on_perf - all power_hvx_on calls now call this function halide_hexagon_remote_power_hvx_on_mode(): - call halide_hexagon_remote_power_hvx_on_perf - change halide_hvx_power_normal to halide_hvx_power_nominal - compute low and nominal modes from max_mips & max_bus_bw - guarantee minimum values for max_mips & max_bus_bw - pass scalars, not halide_hvx_power_perf_t struct Former-commit-id: 35d1db38bf9ae15e011cc442ec118681d278955d 08 December 2016, 21:10:23 UTC
73f176a Merge branch 'master' of ssh://github.com/halide/Halide Former-commit-id: 2124b52cfca852b598ebf176e4c0b3b2a5ed5861 08 December 2016, 19:30:01 UTC
5493c0f Avoid naked used of buffer_t in infer_input_bounds Former-commit-id: 1dd989c291f56ab027b3a356811fd9a77aefbef0 08 December 2016, 19:29:53 UTC
15816ca Merge pull request #1669 from parkhc/refactor-runtime-build Refactoring src/CMakeLists.txt Former-commit-id: 401bea4650419bbb0d23184cebca7219f821293a 08 December 2016, 17:47:20 UTC
a8b2517 Merge pull request #1668 from halide/stubby Add more variants to stubtest Former-commit-id: a7db4d5316b7cca6d7edff257b2eddc51f2d4bb5 08 December 2016, 01:29:40 UTC
9872c36 Add more variants to stubtest Add both simple and array inputs; rationalize names accordingly. Former-commit-id: 47c3c80adff70a0d9c74047e614acacc4712b77b 07 December 2016, 23:58:37 UTC
84b7196 Add missing code Former-commit-id: 804776807d0895bcde12f751adea1342ece2de94 07 December 2016, 22:55:39 UTC
5954141 Formatting tweaks Former-commit-id: e5828b3e9f97c33c9eea8f03d261ff3b376d577f 07 December 2016, 22:37:34 UTC
b656537 Cleanups Moved the allocation fragmenting logic into Halide proper Made it a generator and added a makefile Former-commit-id: 844d969e8f5df401857ff866814a79f82ad0521f 07 December 2016, 22:33:14 UTC
9d002a1 Testing branch for fast cuda mat mul Former-commit-id: 92b3c1f42735f64e03cc695c3a064c7f3a6d700d 07 December 2016, 21:24:32 UTC
7552c23 Merge pull request #1666 from halide/llvm Update llvm error handling Former-commit-id: 5b3a8da1ee21e636207a2cec168748e04b7a218b 07 December 2016, 18:57:43 UTC
9c89eae Refactoring src/CMakeLists.txt Using macro for runtime module generation. Former-commit-id: df4ed1ed509be003327d9584eeddbf1581a134dc 07 December 2016, 18:52:52 UTC
4186661 Fix Part 2 Former-commit-id: e84d1d2d2af4071e960f72f8814d4633060906f8 07 December 2016, 02:02:37 UTC
a1822ec Update llvm error handling LLVM >= 4.0 should prefer explicit llvm::expectedToErrorOr(). Former-commit-id: a1cfa8b592aec25a8068419d60c2fff1af6c37e2 07 December 2016, 01:51:36 UTC
6ca21ab Add missing consts Former-commit-id: 5f873c2c85bb1ad9f8c77e79bba961a0fecc6217 07 December 2016, 01:33:48 UTC
db46cc5 Fix realizing into multiple existing buffers The natural way to realize into multiple existing buffers currently is to put them in a temporary Realization: f.realize({im1, im2}); This doesn't do precisely you want (it used to, before the Buffer change). The Realization constructor makes a copy of those buffer objects that it can own, so you'll lose any mutation of the metadata (e.g. device allocations). It's also different to this: f.realize(im1); which takes im1 by reference and updates the metadata This PR adds a new realization-like class that collects an array of mutable references to existing buffers. It's a temporary type solely used for passing groups of buffers to realize and infer_input_bounds. It changes the meaning of f.realize({im1, im2}) to take im1 and im2 by reference, the way it used to be. Former-commit-id: 73ee0d68d2f0d9b0b36631cf18c70fc6979111f6 07 December 2016, 01:24:57 UTC
095adc1 Add require() operator require() is essentially a runtime assertion; it verifies that an Expr satisfies a precondition (terminating via halide_error() if it does not). Since this inserts runtime checks, it needs to be used with caution: inserting into an inner loop could severely degrade performance. The intended use is for: — preconditions on Param<> inputs that need failure rather than silent clamping — debugging, in situations where print()/print_when() aren’t adequate — in the future: unit testing, via enhancements to Generator Former-commit-id: fd348df611db756d9ede2e625768108037ed6611 06 December 2016, 19:39:42 UTC
e7b04e5 Reformat comment into Doxygen comment for file. Former-commit-id: f651d51d97b75f12ba68f1cbfca914724136d121 06 December 2016, 09:08:07 UTC
efeb9c6 Merge pull request #1663 from kgnk/readd_halide_image_h Re-add halide_image.h Former-commit-id: 743b96c37a36b08494d42b737c7a782c0abf7098 06 December 2016, 09:02:23 UTC
b396d85 add comment explaining use Former-commit-id: e1db28d4762d6e2d75ff53354606a5500e48cdb1 06 December 2016, 07:09:18 UTC
1cce424 readd tools/halide_image.h Former-commit-id: e222c5e47034c2c3ea4322ff293caaf04596d831 06 December 2016, 01:10:04 UTC
0f1d237 Merge remote-tracking branch 'refs/remotes/halide/master' Former-commit-id: 8bb2148917f777f811e2045f3036ff6ce508b778 06 December 2016, 00:44:54 UTC
5931768 Update Metal demo to match runtime declarations for halide_metal_acquire_context and halide_metal_release_context. Update to support all ipad orientations. Former-commit-id: 20e0e00ca4e16fab48ce7b462fa669add9a54253 06 December 2016, 00:16:10 UTC
back to top