swh:1:snp:2c68c8bd649bf1bd2cf3bf7bd4f98d247b82b5dc

sort by:
Revision Author Date Message Commit Date
564aaaf update 07 December 2017, 21:45:16 UTC
792240b adding extract kernels to makefile 07 December 2017, 21:45:16 UTC
5afdea7 add extract kernels 07 December 2017, 21:45:16 UTC
79e3607 refactor device kernel module 07 December 2017, 21:45:16 UTC
a589b0c Change log failure to debug from error, to avoid hiding useful error information. Former-commit-id: 1a2ad70f6f243d3c8a85d8b0790425f055580e95 01 July 2016, 16:54:24 UTC
7724423 Fix hexagon runtime thread pool Former-commit-id: 648f9cd3c65bf821f77868ac38d30bf0b64bbf32 30 June 2016, 19:40:14 UTC
342c5b1 Hide the condition variable declarations because we don't provide definitions on every platform Former-commit-id: d49cd779c9c7d8f4424df9edd280ea71608af541 30 June 2016, 18:28:16 UTC
df71b6e Merge pull request #1354 from halide/llvm_archive_change Track LLVM API change. Not sure if this is the right way to do this. Former-commit-id: d8ae459c2847ec1e6e7c00d7a8bcc6977ae55376 30 June 2016, 16:53:24 UTC
c9c5004 get camera_pipe working nested #if statements are very confusing Former-commit-id: 4813152b1d40dfb7d808282e4d13b233efc27cc8 30 June 2016, 16:52:27 UTC
5763ef5 Merge pull request #1342 from halide/device_zero_copy_buffers Device zero copy buffers Former-commit-id: 37e2ae595fa3d831b54de00468810fb2574c384c 30 June 2016, 02:39:29 UTC
4c097c9 Track LLVM API change. Not sure if this is the right way to do this. Former-commit-id: c68f244cd7ad4e482a934ef06088bb7085ded49f 30 June 2016, 01:33:58 UTC
3f49ba3 Merge pull request #1351 from halide/hexagon_trivial_changes add Hexagon to supports_device_api and all_device_apis Former-commit-id: c1afb6e18504f62a14efb5fb064678bc4f2b2e51 30 June 2016, 01:04:35 UTC
139b4f7 Fix error where stale map entry could be accessed due to mutation rewriting part of the map. Former-commit-id: 3158cc0029b2741e6610fde234adb88ae69d8d55 29 June 2016, 23:35:44 UTC
7a3bdd7 moved invalidate_cache to func::hexagon Former-commit-id: 7455dc732104cb5cfe330db8d6a9e8921d4f2d61 29 June 2016, 23:10:59 UTC
aab77b1 add Hexagon to supports_device_api and all_device_apis Former-commit-id: 26c869f7a09c792b2d4c469f44b9347c83f09709 29 June 2016, 22:52:31 UTC
9dc160f Merge pull request #1350 from halide/hexagon_malloc_cache Added preallocated buffer to speed up halide_malloc/free Former-commit-id: b87df1163b406ed7e5c22941592bf185ec883dc9 29 June 2016, 19:46:31 UTC
ea6c4f1 changed num buffers to 10 Former-commit-id: 49f5da82fcab57132be7c4822d71de3928fe66f9 29 June 2016, 19:43:33 UTC
12c6acc Added preallocated buffer to speed up halide_malloc/free Former-commit-id: be14364e9e45b42f5fddd0a29061f873651ea4e5 29 June 2016, 18:33:37 UTC
1bd8020 Merge branch 'master' of ssh://github.com/halide/Halide Former-commit-id: 3dea55e67ffc3526c992cc80185cd3ea907e540e 29 June 2016, 17:05:22 UTC
4790b3d Expand test_logical to demonstrate hvx bug This catches an apparent codegen bug in hexagon. select on a comparison of a narrow type to mux between two wider types sometimes adds an extra high bit somewhere. E.g. you get 511 as an output instead of 255. Former-commit-id: 3af9af1a531336b812ff04af113949d1ba719b94 29 June 2016, 17:05:05 UTC
d314eec Fix crash on exit in Hexagon simulator. Former-commit-id: 3fa43d52bf48d542c1d70a8e79f1da60ef3984f0 29 June 2016, 15:28:00 UTC
db3e21f Avoid generating non-native vector width code. Former-commit-id: 76616dc8d235091055fdc5bf013b6c689b743567 28 June 2016, 23:59:01 UTC
e53cf1a Turn cache debugging off. Former-commit-id: 08ade4b6d435b7de93879bb480c5a8ec37830ed1 28 June 2016, 23:38:35 UTC
3aec35b Signed/unsigned comparison for cache debugging code. Former-commit-id: e30435cbc696396bab1b768fa69ff02aa3a187ee 28 June 2016, 23:38:08 UTC
9356817 Fix bytes vs. elements logic in memoization cache. Former-commit-id: 9a56e873b76c3323a67576118d11e44b7c73a0a8 28 June 2016, 22:27:23 UTC
70b3cc6 Merge branch 'master' into device_zero_copy_buffers Former-commit-id: 9aece823dec9e3b1a8c5c83234a3d9188cb31242 28 June 2016, 18:32:39 UTC
04fb9f4 Merge pull request #1348 from halide/is-hexagon-available Add halide_is_hexagon_available Former-commit-id: cb043f1cb7c947aecf2e71746b4d26e92557caa2 28 June 2016, 16:43:01 UTC
b012574 Add halide_is_hexagon_available. Former-commit-id: 84cb72954bbb1a37259346f19313ec7fe54c64b5 28 June 2016, 16:12:04 UTC
e34d968 Fix placement of globals Former-commit-id: c52deca584b06356aac8efa3b38a5d69a32ded03 28 June 2016, 15:44:41 UTC
41f9359 Fix double-free in windows thread pool Former-commit-id: c8e927d4db000e3e3c9475175dce744f47ceb2e3 28 June 2016, 15:42:16 UTC
24aaf60 Merge branch 'master' of ssh://github.com/halide/Halide Former-commit-id: a1251543849440393d15584810b5959d8e4ac3ff 28 June 2016, 15:37:37 UTC
af8f833 cleanup -> destroy Former-commit-id: 126cfd0185e87ecb2f3489727f688610409e6946 28 June 2016, 15:37:18 UTC
c385f1c Merge pull request #1347 from halide/hvx-option pass -mhvx as its not the default on tip Former-commit-id: a5b7e7470dc4b6da9876fbfb6e6e02b9675e8452 28 June 2016, 15:36:08 UTC
041ea19 Fix return from spawn_thread Former-commit-id: e42091eb71171369a94ca2fe60aa194992f7d27d 28 June 2016, 04:33:05 UTC
e773acb Fix spawn/join on os x Former-commit-id: d123790502b3cd73dded8812b8003207ace3a680 28 June 2016, 04:05:47 UTC
72b695c spawn thread doesn't take a user context Former-commit-id: e40210b5f02b49ed6a481be5773c9b2675b0f1fa 28 June 2016, 03:36:36 UTC
0179439 Add host_cpu_count for windows Former-commit-id: a492f2364c40a6f3a52b04d45fd5858155d70406 28 June 2016, 03:29:38 UTC
c2f5e94 Fix mutex_destroy name on os x Former-commit-id: f83fbafe5e7f1c6ba8bedfc0b99e9bf0ced73cbe 28 June 2016, 03:06:45 UTC
a042c29 Merge branch 'master' of ssh://github.com/halide/Halide Former-commit-id: e9193f311d8a66922afc3e59d110779292d08474 28 June 2016, 03:04:56 UTC
b6e697e Fix exponential behavior in solver Former-commit-id: 845ac34ddbb211a801be71dfc542117bd11a1505 28 June 2016, 03:04:50 UTC
2c49795 pass -mhvx as its not the default on tip Former-commit-id: f9e339ab1f600cc754aedf910f1b5861e38c7b4b 28 June 2016, 02:57:52 UTC
7ad22a0 Merge branch 'master' into device_zero_copy_buffers Former-commit-id: adb97868445bb612d971879cfc2cb280253e5cde 28 June 2016, 01:21:57 UTC
aea468a Switch message rinted when changing device interface in device_malloc or device_and_host_malloc to an error rather than a debug print. (Shouldn't change behaviro as it returned an error before after printing the debug message.) Former-commit-id: 3babbe89183fd9b257c2ad8da34c7b2bfc760d6a 28 June 2016, 01:20:21 UTC
26f3dd9 Merge pull request #1332 from halide/div_mod_hexagon Div mod hexagon Former-commit-id: 6fcc504724a6e8cfb90ae102549bb533675d1ddd 28 June 2016, 00:12:48 UTC
20fce75 Merge branch 'master' of https://github.com/halide/Halide into div_mod_hexagon Former-commit-id: 8d03e733e83846a59e7466788db39073d4157897 28 June 2016, 00:10:33 UTC
292d3d1 Merge branch 'master' into device_zero_copy_buffers Former-commit-id: 98a765184e81b5ec9bf5072df41459ab7048f5ac 27 June 2016, 23:29:24 UTC
324b457 Enable zero copy support for Hexagon host. Former-commit-id: 61e7057cb098914705c570096b4fe9bd772ead4a 27 June 2016, 23:27:20 UTC
f38d0fd Rename cuda_opencl_shared.h to device_buffer_utils.h as it is not used in a number of other GPU API runtimes. Move buf_size support into new device_buffer_utils.h. Make device_copy memcpy based copying into a common routine. Former-commit-id: 41ef7831e04fc2bddfeb9fbf5dec44d75097461c 27 June 2016, 23:20:38 UTC
c165cb0 Merge pull request #1341 from halide/thread_pool_unification Unify windows and linux thread pools Former-commit-id: fb8f562aa4036c6d94af2cc13591d4bd0c41e5c3 27 June 2016, 23:08:58 UTC
f827e7b Add div and mod library symbols to the table of known symbols Former-commit-id: 60d0852c5dd62f60fa666b48088d11b256c8efd4 27 June 2016, 22:09:43 UTC
34e39e9 msvc 2013 doesn't like constexpr Former-commit-id: fee5b9e2f554f6a07af38f8f821f4fea8fcbc70c 27 June 2016, 22:04:48 UTC
b6e1e07 Do not get the whole libgcc archive Former-commit-id: 48823073555aa22eed01bf45d9b1d3faffb5a42c 27 June 2016, 21:51:04 UTC
14a7a78 Fix issues with trying to do a combined malloc for cases where there are If/Then/Else blocks and both sides do not use the same GPU API. The above results in device_malloc calls being left in, thought they should be entirely redundant when a device and host malloc has already occured. Change "may_be_zero_copy" naming to "device_and_host". Former-commit-id: b73c9def2d3c9a5e497c7595c3001f4313875d86 27 June 2016, 21:44:06 UTC
b9e9d89 Merge branch 'master' into div_mod_hexagon Conflicts: test/correctness/div_mod.cpp Former-commit-id: 5ded0fd48661ef7a087efd9f33cff6c1c601b194 27 June 2016, 21:37:23 UTC
9576e66 Comments, anon namespaces Former-commit-id: 2cccb4dc10e5ec1f700a514af2a840ac68a5dd2b 27 June 2016, 20:40:00 UTC
b6da184 Add some comments about the locks Former-commit-id: f990cbdd016c190e94f199c6d497ef5e30a59d18 27 June 2016, 20:30:44 UTC
f6fff4b The master thread should not drop and immediately reacquire the lock performance_inner_loop_parallel 650ms -> 570ms Also speeds up local laplacian very slightly. Former-commit-id: 0336e573a965cbf4dc75dbce4cf0c75dcf01608e 27 June 2016, 20:27:03 UTC
809e931 Delete commented-out code Former-commit-id: 508391a6a3ed1f35a97abdf0bd1d6f52573e0c83 27 June 2016, 19:54:37 UTC
cf16560 Fix clang build with an enum class verbal lgtm from Steven Former-commit-id: d592bad8c828b8f783359dac8f3c7d236fd95238 27 June 2016, 18:41:14 UTC
a8fc41d Remove global variable and make local instead Former-commit-id: bfe90e1dcee9a7d6930ac85e844c7d706cbc39a0 27 June 2016, 17:13:45 UTC
d56dec7 Merge pull request #1343 from halide/revert-1322-dumber_type_info Revert "Make type_info not depend on addresses of template instantiations" Former-commit-id: b2f4460d9263333fd4a9be72ae1de3f225e2fd80 27 June 2016, 16:55:37 UTC
017e4a2 Revert "Make type_info not depend on addresses of template instantiations" Former-commit-id: d7c404c1f3a880d3c76c4ec9d73b4ef3ba9c13f5 27 June 2016, 16:55:23 UTC
aa58821 Merge branch 'master' of https://github.com/halide/Halide Former-commit-id: 35c43f3a6eb2f0ae78825e107da9d25327222eb5 27 June 2016, 16:06:37 UTC
3b2f278 Merge branch 'master' into device_zero_copy_buffers Former-commit-id: 9005ec6338e266094517d7c0dbcac3689f384c18 26 June 2016, 21:29:03 UTC
4eac43e Add new file that was forgotten. Former-commit-id: b3d410c04e682cdb096bc1211544b9cf6a7186bf 26 June 2016, 21:26:28 UTC
dfcad80 Add support for combined "zero copy" buffer_t allocation. Many GPU APIs, expecially on mobile devices, allow sharing memory between the CPU and the CPU. This change allows the GPU runtime to ustilize this mechanism to allocate both the host and dev parts of a buffer_t at once. This only affects internall allocations. Input and output buffers can generally be allocated zero copy by the caller. Only support for Metal is provided in this first cut. Former-commit-id: 12a3a9c165838446432dc0855eb94aef9097367f 26 June 2016, 21:09:11 UTC
bae5577 Misc cleanups Former-commit-id: 33be3754bd51ecd9ff114a9a5c246e324d1d04f7 25 June 2016, 22:39:10 UTC
b73efb9 Got hvx thread pool working Uses the ability to dynamically change the number of working threads in the thread pool to cope with HVX128/HVX64 changes or even nested parallelism that mixes HVX modes (why would you do that?). Added a test for dynamically changing the number of working threads in the thread pool. Didn't require any new thought in the thread pool - turns out it was already possible. I just did some code cleanup in there. Former-commit-id: eac9c4403f6a1a4a8cbcff2d4570c2cf32581c8c 25 June 2016, 22:31:55 UTC
62663ea Merge branch 'master' into thread_pool_unification Former-commit-id: d1805fd1c12930da36cc98234502b82e51ae75f4 25 June 2016, 20:38:35 UTC
18081d9 Add start of HVX thread pool using common header Former-commit-id: 24bc71f378f5555d75cdc2f0268ee94a9c94be68 25 June 2016, 20:37:54 UTC
ae7cc0f misc cleanups Former-commit-id: 5bcf51d4170468805a3340bd6b1bbfebef784c7d 25 June 2016, 20:37:22 UTC
6e776b5 Move thread pool impl to header so it can be shared with HVX Former-commit-id: 99d782f72bd1e005fada8b445fecd250b2975fad 25 June 2016, 20:37:15 UTC
161e5d9 Merge pull request #1340 from halide/fake_dlopen custom dlopen on hexagon Former-commit-id: 4f5ace94b213b4250713138d9ccc95887bfcae7e 25 June 2016, 19:01:52 UTC
e1b21d3 swap order of printf and fail Former-commit-id: 2aab86dd164f6567711824ea489f8bf8fc29543c 25 June 2016, 19:01:15 UTC
4f92d9a More cleanups Former-commit-id: 9daa8b1c0d829afdf4b17c89d84d5a9076948b34 25 June 2016, 02:37:32 UTC
c1dec0d Merge branch 'fake_dlopen' of ssh://github.com/halide/Halide into fake_dlopen Former-commit-id: 8b0836fbf8b82ce5d21c686339fd6f91ef8fdde9 25 June 2016, 02:30:32 UTC
a050ad1 Misc cleanups Former-commit-id: 2be306ebbc98254a7efdcb393bb6f4c0782f995a 25 June 2016, 02:30:26 UTC
b5144c2 Unify windows and linux thread pools Not yet tested on Windows. The generic code is just a copy-paste of the linux code with "pthread_" replaced with "halide_" Former-commit-id: a32488550321d49bf7a22fcc4bddf9849054427a 25 June 2016, 02:14:12 UTC
6b59f33 Merge branch 'fake_dlopen' of https://github.com/halide/Halide into fake_dlopen Former-commit-id: e652c815d22b6a87f2305b61a3d1e843d17d0da5 25 June 2016, 00:47:11 UTC
ba56bfa Merge branch 'master' of https://github.com/halide/Halide into fake_dlopen Former-commit-id: 6a75f39d6a083479f47cd96058cdfa6925fc70b0 25 June 2016, 00:46:27 UTC
108a164 Merge branch 'master' into fake_dlopen # Conflicts: # src/HexagonOffload.cpp Former-commit-id: edbad5d4034c14595c1ec1cf2f40f14ee0f1e940 25 June 2016, 00:45:46 UTC
84ae84d Merge pull request #1322 from halide/dumber_type_info Make type_info not depend on addresses of template instantiations Former-commit-id: 86a87fe48afbc9200008469cb7d3fd4446c4e352 25 June 2016, 00:05:12 UTC
e3fed84 Merge branch 'fake_dlopen' of ssh://github.com/halide/Halide into fake_dlopen # Conflicts: # src/runtime/hexagon_remote/bin/v60/hexagon_sim_remote # src/runtime/hexagon_remote/elf.h Former-commit-id: c1655474c288c55b76ec5c1efdb28e3de1dd36ff 25 June 2016, 00:04:44 UTC
830356d Add PIC support Former-commit-id: 4eb6e89a6dea2502a2302b8a2e833714059845a7 25 June 2016, 00:04:01 UTC
46c3b8d Merge pull request #1337 from halide/srj-hexlib Add a bottleneck for getting Halide host symbols Former-commit-id: a2fd2610a71003c8088d97bd6452158eb0c91f8b 24 June 2016, 23:15:07 UTC
c038785 fix formatting, remove error checking Former-commit-id: f0561afc5bc9831c1786e2644028515d266431a7 24 June 2016, 23:12:56 UTC
93f60ab renamed (again) to halide_hexagon_host_get_symbol Former-commit-id: 78620036121f5aba72485f6f909fcd5ccde2cadd 24 June 2016, 23:00:34 UTC
aa5f3ad Merge branch 'master' into device_zero_copy_buffers Former-commit-id: 4da350b58242b6b79f7f1627e8f11c814eb50769 24 June 2016, 22:54:28 UTC
bfac1f7 Merge pull request #1336 from halide/undo-lsr-comlexity Revert "Set LSR complexity limit back to the LLVM default" Former-commit-id: 40933452496e96bcb8f1c0cf10eaa4596a4729cc 24 June 2016, 22:30:49 UTC
c9424a8 Make relocation verification a warning and not an error. Former-commit-id: 0055f9f81c1bd091074000ebd2960a0f7d1f73a4 24 June 2016, 22:07:32 UTC
f1a4a11 Merge branch 'fake_dlopen' of https://github.com/halide/Halide into dlopen_crash Former-commit-id: 58dd0c27dcdc0a6c128f670f7679c21f2b5a6d5d 24 June 2016, 21:39:46 UTC
d8244bb Clean up qurt_hvx_lock related declarations. Former-commit-id: 43730674783b6e24f66efad2c668aedecb602593 24 June 2016, 21:39:02 UTC
69be45a Use -G 0 to avoid GP relative relocations. Former-commit-id: 3e29b62d62be16d68a9000ef0cb0e972df003f3e 24 June 2016, 21:37:25 UTC
6a17451 Better schedule, that uses malloc. Former-commit-id: 92f3844f1b3dc8ea1e392fa59e0489d434a9bb94 24 June 2016, 21:27:53 UTC
942bc07 Merge branch 'master' into device_zero_copy_buffers Former-commit-id: fb749be46c3dc804ca2a9922454f0694079be20f 24 June 2016, 21:24:06 UTC
8345054 Fix div/mod for Metal to use common support. Previously it was broken for vector types. Limit vector size in div_mod test for Metal to 4. Go ahead and change a bit of Hexagon logic as it is pending in another and this hopefully makes that a bit smoother. Former-commit-id: 57a348d7466fb55aa8ff1d3e85e12a028aff3205 24 June 2016, 21:21:28 UTC
48c1e6e Checkpoint zero copy buffer work. Former-commit-id: 7c692fbe41db076b9516471cfee08bb2b38bd905 24 June 2016, 19:10:42 UTC
c52b508 Verify relocations don't overflow. Former-commit-id: 67af8243165359ce1095348a99d66217d5902a72 24 June 2016, 18:53:54 UTC
e44e6d8 Merge branch 'master' into srj-hexlib # Conflicts: # src/runtime/hexagon_host.cpp Former-commit-id: 816475b22358c73a06b7042c75ddc7b1e496900e 24 June 2016, 18:43:12 UTC
4739897 Renamed to "halide_hexagon_get_support_lib_symbol"; added comment Former-commit-id: d47a2a7790b51f0e983081db1f375b098c97e698 24 June 2016, 18:40:28 UTC
back to top