| 32ab4a9 | Dillon Sharlet | 16 October 2020, 21:17:22 UTC | Merge branch 'master' of https://github.com/halide/Halide into interpret_nn | 16 October 2020, 21:17:22 UTC |
| 2cde234 | Pranav Bhandarkar | 16 October 2020, 19:43:57 UTC | prefer using HVX over HVX_128 | 16 October 2020, 19:43:57 UTC |
| ba278fe | Steven Johnson | 16 October 2020, 18:24:43 UTC | Merge branch 'master' into interpret_nn | 16 October 2020, 18:24:43 UTC |
| 6f52482 | Dillon Sharlet | 16 October 2020, 17:57:45 UTC | Add TODO on optimizing sum_input | 16 October 2020, 17:57:45 UTC |
| 5286a68 | Steven Johnson | 15 October 2020, 20:57:20 UTC | Fix wasm-related glitches in our timing/benchmarking code | 16 October 2020, 17:40:04 UTC |
| 6986c8f | Dillon Sharlet | 16 October 2020, 16:58:59 UTC | Fix boundary condition and optimize convolution. | 16 October 2020, 16:58:59 UTC |
| a9be7f0 | Dillon Sharlet | 16 October 2020, 07:26:20 UTC | Add max pool operator. | 16 October 2020, 07:26:20 UTC |
| b9e1ee5 | Dillon Sharlet | 16 October 2020, 06:33:39 UTC | Fix split of AveragePool. | 16 October 2020, 06:33:39 UTC |
| 2aa8573 | Dillon Sharlet | 16 October 2020, 06:06:02 UTC | Alphabetize ops. | 16 October 2020, 06:06:02 UTC |
| ce6da60 | Dillon Sharlet | 16 October 2020, 06:02:55 UTC | Rename Add4D -> Add | 16 October 2020, 06:02:55 UTC |
| ce8ca6d | Dillon Sharlet | 16 October 2020, 06:02:34 UTC | Basic optimization and cleanup. | 16 October 2020, 06:02:34 UTC |
| 586a1b0 | Dillon Sharlet | 16 October 2020, 04:03:15 UTC | A few more minor cleanups. | 16 October 2020, 04:03:15 UTC |
| 28d0979 | Dillon Sharlet | 16 October 2020, 03:51:24 UTC | Minor cleanups. | 16 October 2020, 03:51:24 UTC |
| 424b810 | Dillon Sharlet | 16 October 2020, 02:43:33 UTC | Fix bugs in convolution test. | 16 October 2020, 02:43:33 UTC |
| 4e72324 | Dillon Sharlet | 16 October 2020, 02:24:37 UTC | Remove unnecessary hexagon complexity. | 16 October 2020, 02:24:37 UTC |
| d2319de | Dillon Sharlet | 16 October 2020, 01:58:47 UTC | Merge branch 'simplify-nested-broadcasts' into interpret_nn | 16 October 2020, 01:58:47 UTC |
| 44eadb6 | Dillon Sharlet | 16 October 2020, 01:55:57 UTC | Add simplification of broadcasted ramp comparisons, and vector reductions of broadcasts. | 16 October 2020, 01:55:57 UTC |
| f9364f0 | Steven Johnson | 16 October 2020, 00:57:38 UTC | Add convolution_test | 16 October 2020, 00:57:38 UTC |
| 16604ae | Pranav Bhandarkar | 16 October 2020, 00:07:35 UTC | Set vector_size to 128. rule out vector sizes that made sense on HVX_64 now that HVX_128 is the only mode for HVX | 16 October 2020, 00:07:35 UTC |
| 8151b77 | Pranav Bhandarkar | 16 October 2020, 00:04:47 UTC | Check only for Target::HVX | 16 October 2020, 00:04:47 UTC |
| 04cd8dd | Pranav Bhandarkar | 15 October 2020, 23:17:17 UTC | Remove hvx_64 and hvx to python bindings | 15 October 2020, 23:17:17 UTC |
| 9299f07 | Dillon Sharlet | 15 October 2020, 22:10:21 UTC | Enable mobilenet_v2_1.0_224_quant.tflite to run. | 15 October 2020, 22:10:21 UTC |
| 4d1a4bb | Pranav Bhandarkar | 15 October 2020, 21:42:15 UTC | Fix bad merge of test/correctness/mul_div_mod.cpp | 15 October 2020, 21:42:15 UTC |
| 13a4eb1 | Pranav Bhandarkar | 15 October 2020, 21:12:26 UTC | Merge branch 'master' into pdb_remove_hvx_v64 | 15 October 2020, 21:12:26 UTC |
| 09f9eda | Pranav Bhandarkar | 15 October 2020, 21:08:34 UTC | [camera_pipe] - In hvx_128 we need 4 threads to saturate hvx with work | 15 October 2020, 21:08:34 UTC |
| 9a96a46 | Pranav Bhandarkar | 15 October 2020, 21:03:53 UTC | Clean up some nonsensical code related to hvx in apps | 15 October 2020, 21:03:53 UTC |
| 609840c | Pranav Bhandarkar | 15 October 2020, 20:58:37 UTC | Look for Target::HVX too, everywhere that we look for Target::HVX_128 | 15 October 2020, 20:58:37 UTC |
| 3316566 | Pranav Bhandarkar | 15 October 2020, 20:38:51 UTC | Remove all definitions of hvx_64 | 15 October 2020, 20:38:51 UTC |
| 805dbd4 | Steven Johnson | 15 October 2020, 19:37:20 UTC | Update Makefile | 15 October 2020, 19:37:20 UTC |
| 9b594ab | Steven Johnson | 15 October 2020, 19:35:13 UTC | Merge branch 'interpret_nn' of https://github.com/halide/Halide into interpret_nn | 15 October 2020, 19:35:13 UTC |
| ff1b237 | Steven Johnson | 15 October 2020, 19:34:52 UTC | Add Makefile, fix warnings | 15 October 2020, 19:34:52 UTC |
| 4a4653c | Dillon Sharlet | 15 October 2020, 19:14:18 UTC | Some comment and naming cleanup. | 15 October 2020, 19:14:18 UTC |
| 574b6bb | Andrew Adams | 15 October 2020, 18:48:38 UTC | Merge pull request #5359 from halide/abadams/less_scalarization_of_vectorized_atomics Handle more types of nested vectorized += without scalarizing | 15 October 2020, 18:48:38 UTC |
| 2554e60 | Pranav Bhandarkar | 15 October 2020, 18:18:05 UTC | fix intrinsic ids | 15 October 2020, 18:18:05 UTC |
| d113aa6 | Steven Johnson | 15 October 2020, 01:46:07 UTC | Initial push | 15 October 2020, 01:46:07 UTC |
| 98b9074 | Andrew Adams | 14 October 2020, 23:37:33 UTC | Merge remote-tracking branch 'origin/master' into abadams/less_scalarization_of_vectorized_atomics | 14 October 2020, 23:37:33 UTC |
| a33b3fc | Pranav Bhandarkar | 14 October 2020, 22:56:41 UTC | remove use of hvx_64 from Target.cpp | 14 October 2020, 22:56:41 UTC |
| 18b704b | Pranav Bhandarkar | 14 October 2020, 22:54:03 UTC | Remove use of hvx_64 from apps | 14 October 2020, 22:54:03 UTC |
| b121015 | Pranav Bhandarkar | 14 October 2020, 22:46:09 UTC | Remove use of hvx_64 from Halide/test | 14 October 2020, 22:46:09 UTC |
| 6371ef9 | Pranav Bhandarkar | 14 October 2020, 22:31:08 UTC | remove the uses of hvx_64 from Halide/src | 14 October 2020, 22:31:08 UTC |
| 048999c | Pranav Bhandarkar | 14 October 2020, 22:15:19 UTC | Remove MAKE_ID_PAIR and IdPair | 14 October 2020, 22:15:19 UTC |
| 2531d11 | Alexander Root | 14 October 2020, 17:24:04 UTC | bounds inference (mod) pessimistic on unsigned 0 interval (#5350) * move conditions to catch unsigned interval >= 1 case first * use is_int_or_uint() | 14 October 2020, 17:24:04 UTC |
| f89060b | Andrew Adams | 14 October 2020, 17:07:39 UTC | Restrict last test-case to x86 | 14 October 2020, 17:07:39 UTC |
| fb0f632 | Alex Reinking | 14 October 2020, 07:29:32 UTC | Add vcpkg instructions to README.md | 14 October 2020, 15:46:37 UTC |
| 9f13e60 | Steven Johnson | 13 October 2020, 17:46:42 UTC | C backend must use memcpy for load/store | 14 October 2020, 15:41:24 UTC |
| 3f06f37 | Steven Johnson | 09 October 2020, 18:38:25 UTC | Update CodeGen_C.cpp | 14 October 2020, 15:41:24 UTC |
| 2d2b651 | Steven Johnson | 09 October 2020, 18:26:18 UTC | Replace concat() with a union approach. | 14 October 2020, 15:41:24 UTC |
| b8d889d | Steven Johnson | 09 October 2020, 17:56:34 UTC | Fix broken concat() | 14 October 2020, 15:41:24 UTC |
| d161660 | Steven Johnson | 09 October 2020, 00:04:03 UTC | Update CodeGen_C.cpp | 14 October 2020, 15:41:24 UTC |
| 7bec6c8 | Steven Johnson | 09 October 2020, 00:03:00 UTC | Fix relops | 14 October 2020, 15:41:24 UTC |
| fe88544 | Steven Johnson | 08 October 2020, 20:39:35 UTC | Update CodeGen_C.cpp | 14 October 2020, 15:41:24 UTC |
| 3b2d424 | Steven Johnson | 07 October 2020, 21:16:32 UTC | Update CodeGen_C.cpp | 14 October 2020, 15:41:24 UTC |
| c63a80a | Steven Johnson | 07 October 2020, 19:14:15 UTC | Update CodeGen_C.cpp | 14 October 2020, 15:41:24 UTC |
| 6e76f2e | Steven Johnson | 07 October 2020, 19:10:24 UTC | Rewrite internal glue code for vectors in C++ backend. The existing approach (wrapping native vectors in a helper struct) caused too many unnecessary spills. Rewrote so that the underlying Native Vector type is the value being passed around, and improved some of the specializations (with checking of generated code on x64 under clang and gcc). Clang is actually quite good at recognizing patterns and generating appropriate code (at least for x64); gcc is much less so, at least as far as I can tell. | 14 October 2020, 15:41:24 UTC |
| f97c4be | Steven Johnson | 14 October 2020, 15:35:58 UTC | Fix for trunk LLVM | 14 October 2020, 15:40:21 UTC |
| 1e841d6 | Dillon Sharlet | 14 October 2020, 04:27:47 UTC | Merge pull request #5229 from aankit-ca/aankit_long_div Add unsigned long division to CodeGen_Internal. | 14 October 2020, 04:27:47 UTC |
| 474b9e1 | Andrew Adams | 13 October 2020, 23:01:42 UTC | Use a binary reduction tree for outer iterations Also it's better to vectorize at the native width, for both the new and baseline schedules. New inner loop (which does the same number of multiply-adds, but in a 16x4 tile instead of an 8x8 tile): ``` vmovdqu64 (%rcx,%rdx,8), %zmm5 vmovq (%r14,%rdx,8), %xmm6 # xmm6 = mem[0],zero vpermw %zmm5, %zmm0, %zmm7 vpermw %zmm5, %zmm1, %zmm5 vpermd %zmm6, %zmm2, %zmm8 vpermd %zmm8, %zmm3, %zmm8 vpmaddwd %zmm8, %zmm7, %zmm7 vpbroadcastd %xmm6, %zmm6 vpmaddwd %zmm6, %zmm5, %zmm5 vpaddd %zmm5, %zmm4, %zmm4 vpaddd %zmm7, %zmm4, %zmm4 incq %rdx cmpq $32, %rdx ``` | 13 October 2020, 23:01:42 UTC |
| 21b7492 | Andrew Adams | 13 October 2020, 22:10:29 UTC | Fix print | 13 October 2020, 22:10:29 UTC |
| f029301 | Andrew Adams | 13 October 2020, 21:00:39 UTC | Handle more types of nested vectorized += without scalarizing With this change, in a nesting of vectorized vars in an associative/commutative reduction (e.g. +=), we can now have a reduction var outermost and a reduction var innermost and get good codegen. This is still not fully general - there can only be one pure var in the vectorized stack for it to work. In general is_interleaved_ramp should be an is_tensor_contraction pass that knows how to do clever codegen for those. For the following schedule: ``` prod.compute_at(result, x) .vectorize(x) .update() .split(r, ro, ri, 8) .split(ri, rio, rii, 2) .reorder(rii, x, rio, ro) .vectorize(x) .atomic() .vectorize(rio) .vectorize(rii); ``` We get the following IR: ``` let t2262 = (int32x32)vector_reduce(Add, (int32x64(shuffle((int16x15)p10_im_global_wrapper$0[ramp(0, 1, 15)], 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14))*int32x64(shuffle((int16x8)p11_im_global_wrapper$0[ramp(0, 1, 8)], 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7)))) f58[ramp(0, 1, 8)] = slice_vectors(t2262, 8, 1, 8) + (slice_vectors(t2262, 0, 1, 8) + (slice_vectors(t2262, 16, 1, 8) + (slice_vectors(t2262, 24, 1, 8) + f58[ramp(0, 1, 8)]))) ``` The vector_reduce is from rii, and the sum of slice_vectors is from rio. The generated asm (for avx512) is: ``` leal (%rbx,%rcx), %edx movslq %edx, %rdx subq %r14, %rdx vmovdqu (%r15,%rcx,2), %xmm6 vbroadcasti64x4 (%r12,%rdx,2), %zmm7 # zmm7 = mem[0,1,2,3,0,1,2,3] vpermw %zmm7, %zmm0, %zmm8 vpermw %zmm7, %zmm1, %zmm7 vpermd %zmm6, %zmm2, %zmm6 vpermd %zmm6, %zmm3, %zmm9 vpmaddwd %zmm9, %zmm8, %zmm8 vpermd %zmm6, %zmm4, %zmm6 vpmaddwd %zmm6, %zmm7, %zmm6 vextracti64x4 $1, %zmm6, %ymm7 vextracti64x4 $1, %zmm8, %ymm9 vpaddd %ymm5, %ymm8, %ymm5 vpaddd %ymm6, %ymm5, %ymm5 vpaddd %ymm5, %ymm9, %ymm5 vpaddd %ymm7, %ymm5, %ymm5 addq $8, %rcx cmpq $128, %rcx ``` The pmaddwds are from rii, and the vpaddds are from rio. This is 3.5x faster than the best schedule that only vectorizes the pure var, and about 2x faster than the best schedule that only vectorizes an unsplit reduction variable. | 13 October 2020, 21:00:39 UTC |
| 47a5a44 | Andrew Adams | 12 October 2020, 23:32:57 UTC | min and max in the algorithm was confusing loop partitioning Because if any likely tag at all existed on a side of the min/max, even if captured, the other side wasn't getting mutated. This should only happen for uncaptured likelies, where simplifications in the unlikely path are irrelevant. | 13 October 2020, 16:03:51 UTC |
| e8430ea | xndcn | 13 October 2020, 12:03:41 UTC | Append missing newline "\n" for camera_pipe usage. | 13 October 2020, 16:00:06 UTC |
| 57cac82 | xndcn | 13 October 2020, 10:42:55 UTC | Remove redundant last ")" for HalideTraceViz usage | 13 October 2020, 16:00:06 UTC |
| b83cad5 | Ankit Aggarwal | 13 October 2020, 09:15:18 UTC | Add calls to simplify | 13 October 2020, 09:15:18 UTC |
| a5d91d5 | Dillon Sharlet | 13 October 2020, 03:40:45 UTC | Merge pull request #5349 from halide/abadams/nested_vectorization_compile_time_regression_fix Fix compile-time regression relating to nested vectorization change | 13 October 2020, 03:40:45 UTC |
| 396cf0c | Shoaib Kamil | 13 October 2020, 00:04:58 UTC | Remove erroneous sentence in comment about GCD (#5354) Grand Central Dispatch isn't used anymore; this comment is outdated. | 13 October 2020, 00:04:58 UTC |
| 13ae382 | Andrew Adams | 12 October 2020, 23:02:02 UTC | Fix comment for halide_error_code_device_dirty_with_no_device_support (#5352) | 12 October 2020, 23:02:02 UTC |
| facb69d | Andrew Adams | 12 October 2020, 20:16:43 UTC | Fix for unbounded lanes | 12 October 2020, 20:16:43 UTC |
| 50c61b9 | Andrew Adams | 12 October 2020, 20:04:10 UTC | Fix compile-time regression relating to nested vectorization change The min_lane expression could grow very large, and required simplifying once per lane | 12 October 2020, 20:04:10 UTC |
| 96772ae | Dillon Sharlet | 12 October 2020, 18:09:10 UTC | Stronger simplification of fused ramps (#5343) * Improve simplifier for fused ramp expressions. * Add test checking that fused ramps simplify away. * clang-format * Use custom lowering pass instead of file system to check for mod. Co-authored-by: Dillon Sharlet <dsharlet@gmail.com> | 12 October 2020, 18:09:10 UTC |
| cc34401 | Andrew Adams | 10 October 2020, 00:45:52 UTC | Better simplification and codegen for nested vectorization (#5325) * Improvements to nested vectorization simplification and codegen * Add test for nested vectorization perf * Add detection for transpose-shuffles * more ramp-of-ramp simplification * Better pmaddwd recognition for VectorReduce in x86 backend * Add pmaddwd for avx512 | 10 October 2020, 00:45:52 UTC |
| fa54197 | Steven Johnson | 08 October 2020, 21:23:52 UTC | Remove the ADD_[U]INT64_T_SUFFIX macros Modify codegen for C-like backends to just emit integer constants with the correct suffix for the backend, rather than wrapping in a giant macro; the macro approach worked but lordy was it painful to read. | 09 October 2020, 16:24:11 UTC |
| a1c9d89 | Steven Johnson | 08 October 2020, 17:39:19 UTC | Fix for trunk LLVM | 08 October 2020, 22:50:01 UTC |
| 3db6e2c | Ankit Aggarwal | 08 October 2020, 21:10:40 UTC | Add braces around while | 08 October 2020, 21:10:40 UTC |
| a189fd4 | Andrew Adams | 07 October 2020, 18:36:52 UTC | Merge pull request #5335 from jlaxson/gpu-test-workgroup-size Reduce GPU Tile Size in Tests | 07 October 2020, 18:36:52 UTC |
| 018d70e | Ankit Aggarwal | 07 October 2020, 06:37:14 UTC | Modify mod_round_to_zero to use remainder from long_div | 07 October 2020, 06:38:48 UTC |
| 9d48b63 | Ankit Aggarwal | 06 October 2020, 19:25:00 UTC | Run clang-format | 06 October 2020, 19:25:00 UTC |
| f99018c | Ankit Aggarwal | 06 October 2020, 19:15:18 UTC | Add long div/mod to CodeGen_Internal. | 06 October 2020, 19:15:18 UTC |
| b1fd538 | Steven Johnson | 05 October 2020, 22:41:32 UTC | Make C++ backend requirement for C++11 explicit Some C++11 features had crept into our C++ backend codegen; make this explicit and check for the correct version at the top of the generated file. (Then remove the stray checks for C++11 version elsewhere.) | 06 October 2020, 16:56:23 UTC |
| d012df7 | Andrew Adams | 06 October 2020, 16:35:49 UTC | Merge pull request #5331 from rootjalex/master fix bounds inference bug for bounded interval / unbounded interval | 06 October 2020, 16:35:49 UTC |
| 2d15069 | Andrew Adams | 06 October 2020, 16:34:55 UTC | Merge pull request #5334 from halide/abadams/braces_around_statements Require braces around if/while bodies. | 06 October 2020, 16:34:55 UTC |
| 8c12c67 | John Laxson | 06 October 2020, 02:00:54 UTC | Reduce GPU Tile Size | 06 October 2020, 02:00:54 UTC |
| ecad269 | Andrew Adams | 05 October 2020, 23:56:18 UTC | Use switch statement instead of if sequence | 05 October 2020, 23:56:18 UTC |
| 325490b | Andrew Adams | 05 October 2020, 23:53:09 UTC | Require braces around if/while bodies. | 05 October 2020, 23:53:09 UTC |
| eb279fc | Alexander Root | 05 October 2020, 22:18:10 UTC | update comment on bounds bug | 05 October 2020, 22:18:10 UTC |
| 408a0e2 | Alexander Root | 05 October 2020, 21:45:48 UTC | fix bounds inference bug for bounded interval / unbounded interval | 05 October 2020, 21:45:48 UTC |
| 68f66fe | Shoaib Kamil | 05 October 2020, 17:55:33 UTC | Fix 32-bit Windows vcvars command (#5330) `x64_x86` = build using 64-bit compiler, targeting 32-bit x86 (https://docs.microsoft.com/en-us/cpp/build/building-on-the-command-line?view=vs-2019). We had it backwards before. | 05 October 2020, 17:55:33 UTC |
| 86fa657 | Steven Johnson | 02 October 2020, 22:02:38 UTC | registerPassBuilderCallbacks is only available in LLVM 12+ | 04 October 2020, 18:16:12 UTC |
| 2d3ebec | Ankit Aggarwal | 01 October 2020, 00:38:14 UTC | Formatting. | 04 October 2020, 18:16:12 UTC |
| ea583f3 | Ankit Aggarwal | 26 August 2020, 18:20:45 UTC | Add TargetMachine::registerPassBuilderCallbacks Add call to TargetMachine::registerPassBuilderCallbacks to allow targets to add passes to the pass pipeline using the New Pass Manager | 04 October 2020, 18:16:12 UTC |
| fa9da79 | Steven Johnson | 03 October 2020, 17:16:31 UTC | Merge pull request #5324 from halide/srj/inject_buffer InjectBufferCopiesForInputsAndOutputs should check for unexpected Call nodes. | 03 October 2020, 17:16:31 UTC |
| 3d19071 | Steven Johnson | 03 October 2020, 04:18:10 UTC | Merge pull request #5315 from halide/srj/abort Kill halide_abort() | 03 October 2020, 04:18:10 UTC |
| fa4d11b | Steven Johnson | 03 October 2020, 00:33:29 UTC | InjectBufferCopiesForInputsAndOutputs should have an assertion for Call nodes | 03 October 2020, 00:33:29 UTC |
| 908e626 | Steven Johnson | 02 October 2020, 22:14:00 UTC | Kill halide_abort() This was added long ago as an attempt to work around issues with the Windows Debug runtime (in which calling `abort()` would produce an "Abort, Retry, Ignore" dialog). We no long do debug builds of any sort on our buildbots, so let's lose all this mess to simplify our world a bit. | 02 October 2020, 22:14:00 UTC |
| a1206b5 | Steven Johnson | 02 October 2020, 21:11:49 UTC | Merge pull request #5322 from halide/srj/runtime-warn Make sure the runtime compiler settings in CMake match those in Make | 02 October 2020, 21:11:49 UTC |
| bacd284 | Steven Johnson | 02 October 2020, 18:47:48 UTC | Make sure the runtime compiler settings in CMake match those in Make Mainly, we weren't setting any of the warning flags, so CMake builds were more forgiving than Make. | 02 October 2020, 21:11:28 UTC |
| 14b567e | Steven Johnson | 02 October 2020, 18:49:18 UTC | Merge pull request #5312 from halide/vksnk/align_loads Don't try to align loads if alignment is not divisible by the size of the load | 02 October 2020, 18:49:18 UTC |
| ac56ec9 | Steven Johnson | 02 October 2020, 16:56:02 UTC | Merge pull request #5308 from halide/build/shared-llvm-fix Fix linking to shared LLVM. | 02 October 2020, 16:56:02 UTC |
| ff5c2ad | Steven Johnson | 02 October 2020, 01:07:11 UTC | Merge branch 'master' into vksnk/align_loads | 02 October 2020, 01:07:11 UTC |
| 08ab4a5 | Steven Johnson | 01 October 2020, 22:35:18 UTC | Merge pull request #5321 from halide/docs/readme-homebrew Add package manager info to README.md | 01 October 2020, 22:35:18 UTC |
| 2768ee3 | Alex Reinking | 01 October 2020, 21:52:23 UTC | Add package manager info to README.md | 01 October 2020, 21:52:23 UTC |
