https://github.com/halide/Halide

sort by:
Revision Author Date Message Commit Date
d0dceeb Comment out debug print Former-commit-id: 9cb8915733bf54df4e2da25f8b3e1eeb623c645c 05 February 2017, 22:57:14 UTC
5c0d610 Revert test Former-commit-id: 1da1b3c15f04de30d6aed35cd27ef176467c2c44 05 February 2017, 22:55:53 UTC
af050aa Fix hexagon load scalarization Former-commit-id: e691ba6f5e0dcdd2b14bbdb7cc6b41764f82049a 05 February 2017, 22:53:50 UTC
6769a10 Hexagon should scalarize load with stride not equal to one Former-commit-id: a8dad8f0f4c37faa33cfd65cc0aaa5e8c9a27fae 05 February 2017, 20:25:44 UTC
6b6d52a Revert -fno-var-tracking-assignments, doesn't exist on some compilers. Former-commit-id: 2d88c39006c25a27d4d07093d57e99ebec198897 04 February 2017, 20:43:46 UTC
aac4c0e Add missing include. Former-commit-id: 4f97fd1c115937f7c2d049b33896abae4e22992a 04 February 2017, 20:13:41 UTC
2d37ee3 Fix build for Hexagon. Former-commit-id: 26adc176523ca1ec7fc9513b1c7e40d5b6c3a06b 04 February 2017, 20:06:15 UTC
749bcb6 Fix broken headers with file paths as function names. Former-commit-id: a737e1890b21f7132bcde45241c14a6e87c44533 04 February 2017, 20:05:35 UTC
5a7b358 Fix race condition around setting LLVM arguments. Former-commit-id: 2cd2723b4aaa51aea36845b27135548669e42632 04 February 2017, 20:05:08 UTC
c40836b Turn off -fvar-tracking-assignments for simd_op_check Former-commit-id: d0956d412fa1d37e18690c6120f2e6bcf9c072e1 04 February 2017, 20:04:01 UTC
689b33a Fix #include and warnings in halide_test_dirs.h Former-commit-id: ff659ac6dc1c04669b016b0e5dd8b8c4c1578007 03 February 2017, 19:39:42 UTC
01f4d46 Merge pull request #1801 from halide/specialization_on_compound_condition Beef up specialization on non-trivial boolean exprs Former-commit-id: eda0d205b96a1e02ae41073b67432cdbdca94345 03 February 2017, 17:48:10 UTC
e953d6b Limit thread usage of vector_cast tet Former-commit-id: 6a7b7266a0f03977327b171e025ab8d89076db05 03 February 2017, 04:10:13 UTC
c787c2d More explicit test Former-commit-id: df02a5827aa4efdfddfe5642562dce0d5c4e3561 03 February 2017, 00:00:51 UTC
256b9a0 Fix comments Former-commit-id: 6b4f3f2cc05b6383a7d0f2d46333a2b76cf58eef 03 February 2017, 00:00:42 UTC
20d9ce5 More fixes for older llvms Former-commit-id: a3cb5920a26b02f6144cd4fb02eb3dd9b256e601 02 February 2017, 20:55:01 UTC
994563a mingw fix Former-commit-id: baed8814da1a607fea0b6f484dc0c4277c30674a 02 February 2017, 20:43:14 UTC
192cdae Fix for older llvms Former-commit-id: a403f062d337759787199524782a97e523cc260c 02 February 2017, 20:06:15 UTC
daaf21b Merge branch 'master' of https://github.com/halide/Halide Former-commit-id: 8edcd8cd8dcf1d93c89f8492bbce1a9ef4953ef8 02 February 2017, 18:32:17 UTC
4392afc Fix for trunk llvm Former-commit-id: 9eb3888670a9bb5f22ff37d6e40d25e3aeb5a992 02 February 2017, 18:32:06 UTC
1024007 Fix test; remove debugging print Former-commit-id: 75feb687d21fa90d2bdaa8d711ae0c02800d8b91 02 February 2017, 18:01:50 UTC
4168710 Merge pull request #1799 from halide/limit_threads_simd_op_check Limit max threads in simd_op_check Former-commit-id: 298f2a1ca819164535e06b6d2ce17e2d67de0560 02 February 2017, 17:46:25 UTC
9a1fedb simplification rules for checking if a value is in or out of a range Former-commit-id: 6e71cb1403f1d043f244ebc20030f183d3cc7907 02 February 2017, 01:41:50 UTC
fbad2fa Only single-arg version of Buffer ctor needs to be explicit. A little code repetition but better compatibility with existing code. Also fixed the one site affected by the newly explicit ctor. Former-commit-id: f54a2d1649b1efe13bbe5bbfb6a6febc2d294e3b 02 February 2017, 01:25:53 UTC
23051bc Merge pull request #1802 from halide/srj-explicit Add "explicit" to Buffer ctor Former-commit-id: 10a7ea73b88f9415b9850e5f0ce3d0008bd0fa86 02 February 2017, 01:14:57 UTC
44178d3 Add "explicit" to Buffer ctor Otherwise, a T is implicitly convertible to a Buffer<T> (for T != void), e.g.: Buffer<uint8_t> b; const uint8_t u = 0x42; b = u; // this works, but is unlikely to be what you wanted Former-commit-id: e1d5629642e6aedb5db35a155533033e948166fa 02 February 2017, 01:04:17 UTC
4bf313e Beef up specialization on non-trivial boolean exprs Former-commit-id: b3ca6e118978fa645ebd96994c8c5b8ccb015892 01 February 2017, 23:40:02 UTC
536ad2a Add comments Former-commit-id: 84e7191bd6cd5417e4450af2fe67bc406f35efcc 31 January 2017, 22:03:30 UTC
5deb518 Merge pull request #1798 from halide/srj-sched Allow Generator::schedule() to be a std::function member var Former-commit-id: 6d3de3821354fa4551fbc0be8db71a4f3970090e 31 January 2017, 21:33:53 UTC
1eb9ec8 Limit max threads in simd_op_check Former-commit-id: 3b033c840d1835877e08c3db96394a3bb6aba077 31 January 2017, 21:25:22 UTC
970953a Windows fix Former-commit-id: c0855acd70c5ea4b10c282ec1a1d6651943c9bae 31 January 2017, 20:26:23 UTC
dad0661 Allow Generator::schedule() to be a std::function member var The static_assert logic in Generator required schedule() to be an actual member method; now it can also be a member variable (e.g. std::function) that is filled in by the generate() method, to support lambdas for scheduling. Modified pyramid_generator.cpp to use this approach. Former-commit-id: d6f53eb823241e82ac24b1f0b7d08b844927ea22 31 January 2017, 01:06:36 UTC
f41a8d9 Merge pull request #1795 from halide/async Use std::async to speed multitarget builds Former-commit-id: d79e4f36c3d7899f0dc47f2a7f14b044b5b8ea79 31 January 2017, 00:38:57 UTC
f22f37e Remove trailing space Former-commit-id: 4a726168ff7de909d4e7b037edb492cea6d17501 31 January 2017, 00:38:39 UTC
fa4cb7c Revert mistaken checkin Former-commit-id: e883cc33f690e00d19f6648210c38c7b2525c7c5 31 January 2017, 00:37:44 UTC
55f93b2 Merge branch 'master' into async # Conflicts: # src/Module.cpp Former-commit-id: ed0cd9e090d495a7a43ac044d44099eac590502c 31 January 2017, 00:35:12 UTC
68d8c19 Merge pull request #1796 from halide/srj-sfbn Add Module::get_function_by_name() Former-commit-id: 14d9ffe3d86895b54cb39f2c46e68cf5680a97a2 31 January 2017, 00:27:04 UTC
0528c4e Add Module::get_function_by_name() Some code assumes that functions().back() is the ‘public’ fn in a Module, but this isn’t guaranteed and will change soon. Add a function to look up by name and use it in a few relevant places. Former-commit-id: c27b8814e44c5d5cb7fa8a44c7289fbbccdc9a6b 30 January 2017, 19:01:40 UTC
0539163 Merge pull request #1787 from halide/srj-boun Convert boundary_conditions test to use std::async instead of fork() Former-commit-id: dd8aeb09607c9dff40e3fa444037550ba5a17592 30 January 2017, 18:15:17 UTC
44f05c4 Merge pull request #1785 from halide/srj-simd Use std::async to speed mul_div_mod test Former-commit-id: bf8ffafe81c2073d444aa856007964323757a94a 30 January 2017, 18:15:05 UTC
4755fb5 Merge pull request #1788 from halide/srj-ts thread_safety test should use std::thread instead of OMP Former-commit-id: eed4707362495f9cac4e5bed3282f8b521deb6d8 30 January 2017, 18:14:56 UTC
782a76a Merge pull request #1789 from halide/srj-vc Speed up vector_cast test by using std::async Former-commit-id: 81a0667592a271db09c0c9ca45ca1695845ca313 30 January 2017, 18:14:42 UTC
61e3ee1 Merge pull request #1790 from halide/srj-vm Speed up vector_math test by using std::async Former-commit-id: aa0b14e7a5fb5f9904a314039882e50b31a51006 30 January 2017, 18:14:32 UTC
c9c6e29 Merge pull request #1792 from halide/srj-opchk Rework simd_op_check test Former-commit-id: ed8f8bec149a1c83c0429313d1bd214a2cd27443 30 January 2017, 18:14:20 UTC
feff3e3 Use std::async to speed multitarget builds compile_to_target() now executes all calls to Module::compile() via std::async. Speedup depends on machine, OS, number of targets, etc., but the compile_to_multitarget (with two targets) test goes from ~600ms -> ~400ms on my four-core Mac. If I bump that test from 2->5 targets, test goes from ~900ms -> ~425ms. I tried to keep the async usage as conservative as possible, but the possibility for thread-related bugs exists, of course. I’m going to patch this CL into some code at Google for torture testing, as multitarget outputs are much more heavily covered there than in the Halide repo. Former-commit-id: b48100345cee3e0e5f84c156de75f45f8acea1a0 27 January 2017, 23:23:08 UTC
cb1c577 clang_tidy fix Former-commit-id: b506eef0b232be46f46f4e99ccd1792b782144c9 27 January 2017, 22:40:20 UTC
854e12c Add missing registration of filter_metadata_t Former-commit-id: 8b120c652d8cedb75054e82c8033e019f5044e54 27 January 2017, 22:39:07 UTC
946dc8a Merge pull request #1793 from halide/clang_tidy_macro_fix Pacify clang-tidy Former-commit-id: dd07ee1a601f5c435a701f9e00d6175fc15e2b67 27 January 2017, 21:53:51 UTC
96c8f20 OS X build fix Former-commit-id: 1eb4c2ad31e8992acd9c65ddfb3adf0b87a30465 27 January 2017, 20:48:43 UTC
cade249 Windows fix Former-commit-id: a541ab3e6a472767f085a31de448cb6f1360d57d 27 January 2017, 20:47:04 UTC
40f81af Pacify clang-tidy The extern macros define functions in user code, where overzealous clang-tidy linter passes are free to complain about it. We can hopefully shut it up by just following its advice for the macro bodies. Former-commit-id: 93454812fa4e56757722fd8350c1f79dab8e2817 27 January 2017, 20:45:32 UTC
b9d03fd Use a pool of 8 threads. Former-commit-id: 510dd43dc2e77f0fdeef154e0b94e36d7e55b52e 27 January 2017, 19:29:55 UTC
581c4b7 Merge pull request #1784 from halide/srj-test apps/simd_op_check fixes Former-commit-id: 805fb98488e735a9c11193e74985b1a1f1cfe74c 27 January 2017, 19:24:26 UTC
993ba8b Merge pull request #1786 from halide/declare_extern_type_macro Add convenience macro for declaring types Former-commit-id: b4741b731c6d97b76b602ee7a23fb05b94efb49e 27 January 2017, 19:15:23 UTC
59a623c Fix for trunk llvm Former-commit-id: 226fde44f84eb97f11fd5fceb62e52db89ca2433 27 January 2017, 18:08:03 UTC
d5987f3 #define NOMINMAX in halide_test_dirs.h Instead of #undef min/max elsewhere. Former-commit-id: 011d86b7a059996a1695de69cc8627b8b39c0c4e 27 January 2017, 02:02:21 UTC
9ff6ca8 Use std::async() instead of std::thread() It is guaranteed to use a pool-like behavior; to ensure that every task is executed by spawning a new thread, we must explicitly specify std::launch::async. Former-commit-id: 4236c652b3d581ce8b6851efd4653cb815c4ac70 27 January 2017, 01:59:40 UTC
ac4a32d #undef min and max for MSVC Former-commit-id: aecfebbc0a18ea495925c3a133267dc8cd8ee4dd 27 January 2017, 01:44:40 UTC
9795389 Allow simd_op_check on Windows Former-commit-id: bf0c54ee98a5a8a410e928d64958307623a32593 27 January 2017, 01:26:48 UTC
b97c939 Allow cxx_mangling_aottest to build when PTX is disabled Former-commit-id: 6605c1276cd3a7e3697318077904490304a5a3a8 27 January 2017, 01:09:04 UTC
28bb288 Rework simd_op_check test — use std::async instead of fork(), so we can now usefully test on Windows. (On my Mac laptop, this runs slightly faster than the previous fork() solution.) — Refactor the code to be class-based, and move all globals into member vars; this makes it much easier to reason about what is read-only vs read-write during test execution vs test initialization — add an extra, optional cmd line arg to allow the temp files to be output to a particular directory; default this directory to Internal::get_test_tmp_dir(). Former-commit-id: 007af53c60b3355dcd1388a4ac9c7ea1cbb1e96b 27 January 2017, 00:24:31 UTC
8ea0d12 Support unions. Cleanup. Former-commit-id: c4ee2cd15ae20056134adff75abd5b84378b404b 26 January 2017, 23:31:13 UTC
1c3b071 Fix forward decl code Former-commit-id: f5d6cfb1d1d1b809873a3d31cdedc44ef5613752 26 January 2017, 23:31:03 UTC
1a14a9d Merge pull request #1791 from halide/revert-1777-srj-multi Revert "Remove Generator multitarget special-cases" Former-commit-id: 95107064dc10bf81507f56f0cc0ff1dd22d858a1 26 January 2017, 23:13:00 UTC
566fdc4 Revert "Remove Generator multitarget special-cases" Former-commit-id: c915859fe2fb6cb1bfaed8c836b17a30e7498b9e 26 January 2017, 23:04:38 UTC
000ec44 Remove unnecessary wait() Former-commit-id: 05a60a54adab959aa5bbad78c1e19dfff0a53b31 26 January 2017, 22:19:43 UTC
03c728e Remove unnecessary wait() Former-commit-id: 396a27b94a7f249d7bf2c3111e68d5ea771be344 26 January 2017, 22:19:11 UTC
48b8f80 Remove unnecessary wait() Former-commit-id: 8e5d53d27c3c7e1c61a73b595e5f7ed251a7a807 26 January 2017, 22:18:39 UTC
bbea4eb Remove unnecessary wait() Former-commit-id: 8647e537a353e3f2da87d09b6a6fe8893fd981ed 26 January 2017, 22:17:57 UTC
4a08ac1 Speed up vector_math test by using std::async ~4x wall-clock improvement on my laptop. Former-commit-id: 721febcab3f1d3c5d392e42fe9a46002e526babb 26 January 2017, 22:03:58 UTC
cd37a16 Speed up vector_cast test by using std::async ~4x wall-clock improvement on my laptop. Former-commit-id: 2772afddd3fd487337b7c2e419198692f9a7ec74 26 January 2017, 21:54:52 UTC
4e3ccce Merge pull request #1777 from halide/srj-multi Remove Generator multitarget special-cases Former-commit-id: 7fd6af0a18fe3dbab08502c515f9c1de859fa3d4 26 January 2017, 20:56:01 UTC
1ff98a5 thread_safety test should use std::thread instead of OMP OMP isn’t available on all interesting targets, so this test serializes (and effectively hangs) on those. Explicitly spawn threads instead. Former-commit-id: 7c9c363723b2a05ab973a2e3da1e0003467feaa9 26 January 2017, 20:48:53 UTC
76ca51a Tweak comment Former-commit-id: 3f9b5bae37bdbe8d0b6b25e7ca2728c9cb8a2803 26 January 2017, 20:40:58 UTC
d93295a Convert boundary_conditions test to use std::async instead of fork() Allows Windows usage of this test; also cleans up code a bit. (Further de-duping is possible but unsure if worthwhile.) Former-commit-id: bf904bdca87d7264133d6faeb29f0e7c99439891 26 January 2017, 20:40:06 UTC
6d22944 Add convenience macro for declaring types Helps with name mangling in code using define_extern that passes pointers to user types as arguments. Former-commit-id: 237e26bc24470b447024bbb5a8ed6c5adcb174ea 26 January 2017, 20:26:15 UTC
4859d87 Use std::async to speed mul_div_mod test Running tests in parallel speeds up wall-clock execution time of this text ~4x on my laptop. (Driveby: denoise output, re-route errors to stderr) Former-commit-id: 8f9d51f3904c039267ffb0efd07d7097a5dc0fe0 26 January 2017, 20:26:13 UTC
dd9a1fd apps/simd_op_check fixes — driver.cpp uses posix_memalign for OSX compatibility — revise Makefile to put all generated files into a bin/ subfolder, to match practice of all other apps/ Former-commit-id: 0fa635dbf7f0de13748ce2d8acb04cce513cc61c 26 January 2017, 20:05:57 UTC
823a66e Merge pull request #1783 from estollnitz/opengl_select_bug Test cases and fix for issue #1781 Former-commit-id: 409fb0053c53cc2a5e4a891cfe51e71d1d7e6436 26 January 2017, 18:06:57 UTC
19816db Fix double using decl Former-commit-id: 4d9975e29b43c95e7424d0df2d7584019919531d 26 January 2017, 01:02:03 UTC
12d1565 Skip producing cpp for the multitarget Generator test Former-commit-id: ad1121c3acd533712980f15eea116be91e82cec1 26 January 2017, 00:08:41 UTC
63749ed Merge pull request #1774 from halide/hvx-power-opt Reduce power_hvx_on overhead when it isn't necessary Former-commit-id: 7f7875fc44a41845cd54c2d5a2805992823b258b 25 January 2017, 23:31:24 UTC
a628ce7 Fixed #1781. In VaryingAttributes.cpp, CastVariablesToFloatAndOffset was mutating the true and false branches of a select expression but neglecting to mutate the condition. And for ramp expressions, if either base or stride mutates into a float, then both need to be floats. Former-commit-id: 35b346a63439ba34665bed33ae76a9729c3e4656 25 January 2017, 23:26:14 UTC
3709f17 Merge pull request #1780 from halide/lerp_ub_fix Fix signed integer overflow in lerp Former-commit-id: e2ff4aa6b92b38f7b496797381b6c6b933c0ff12 25 January 2017, 23:03:58 UTC
cd6389f Added failing test cases for issue #1781: Internal assertion failures when compiling certain select expressions for OpenGL target. Former-commit-id: 3fde18d359a37115da08aeb0a1185a16f505caf4 25 January 2017, 22:54:40 UTC
3073e0e Add comment Former-commit-id: 78480779485bc6d1df3e1f385a5f46220699ace4 25 January 2017, 22:39:41 UTC
d82aa67 Fix signed integer overflow in lerp Former-commit-id: 92ea983e5d26e542e5f4b7f768c8ec17d16ec824 25 January 2017, 22:15:55 UTC
b945b40 Merge pull request #1779 from halide/mangling_issue Mangling issue Former-commit-id: 8d6a032da54372f3edf5e46b2fa3cb585a8fe7a5 25 January 2017, 18:48:23 UTC
47c91a0 Merge pull request #1778 from halide/srj-nbuf IO<Buffer> should have allowed for dim-only and type-only Former-commit-id: fee84f09e17a6dd274970f56c0cd8b8d29eb9fdd 25 January 2017, 01:30:44 UTC
b831a4f Clean up clowniness of false economy. Former-commit-id: 499970bdf820b88e39e12d357a8cdcc7f2a08529 24 January 2017, 23:49:57 UTC
6b0cb27 c-source output must have an unadorned filename and fail for targets > 1 Former-commit-id: 24434a8b9a3daef75830ae5407a5bd401cf249c9 24 January 2017, 23:42:30 UTC
bf4c2bb Fix issue with special casing of nullptr handle type mapping to void * mangling. On Itanium ABI, this did not enter the Pv entry into the substitution table. On Windows, the wired manglings were just wrong. Added test for the case. Former-commit-id: ec0027cb8a28f7b7e4136d16a358e0f6856b09bb 24 January 2017, 23:34:22 UTC
030a799 Merge branch 'master' into srj-multi Former-commit-id: 1badd7c91a816501a358d53dd489eb42f96d0ee4 24 January 2017, 23:22:18 UTC
4aad304 Merge branch 'master' into srj-nbuf Former-commit-id: f96890fb783e9c065c94121f73aad13086269c17 24 January 2017, 23:17:35 UTC
c25b677 Remove unused const variable to silence warnings Former-commit-id: 9a476bba5d5953868fe174ef98e11f7cf09ff2b6 24 January 2017, 23:17:15 UTC
c870287 IO<Buffer> should have allowed for dim-only and type-only Specifying an Input<Buffer> or Output<Buffer> with one of type or dim (but not both) should have worked, but didn’t. Now it does. Former-commit-id: 3c7ed83d4ef18f21945e50f19d3e86dd465b6392 24 January 2017, 23:16:14 UTC
6569e6b Add user_context to extern mangling test Former-commit-id: 983deb2c0240018dd69be160329242d238980c21 24 January 2017, 22:20:03 UTC
ddcdf86 Revise compile_multitarget to avoid emitting the wrapper code for the case of a single target. This still reduces the separate code-paths for generator, while still achieving wrapperlesssingle-target output. Former-commit-id: 95f6d01b71c2e9162e1d02fc19e1664d77fcc4ea 24 January 2017, 22:03:56 UTC
44fa1c5 Merge branch 'master' into srj-multi Former-commit-id: 8b94418468c1998eb84ec758abb313764bda60fc 24 January 2017, 21:30:27 UTC
41b9aec Merge pull request #1775 from halide/buf Minor code simplification in Module.cpp Former-commit-id: 6e05abeec838a5af43b78ff7637e5caa128de107 24 January 2017, 20:22:28 UTC
back to top