Fix TVM compilation with USE_LLVM=OFF #194

* [µTVM] Specify loader for yaml.load Specify the loader to be used by yaml.load as the current form used without specifying explicitly a loader is deprecated since PyYAML 5.1 and will throw a noisy warning. For details, please see: https://github.com/yaml/pyyaml/wiki/PyYAML-yaml.load(input)-Deprecation Signed-off-by: Gustavo Romero <gustavo.romero@linaro.org> * [µTVM] Avoid using tvm.target.create Avoid using tvm.target.create as it's deprecated and use tvm.target.Target directly instead. Signed-off-by: Gustavo Romero <gustavo.romero@linaro.org>

* Added Ops * Regular * Remove copy * Remove copy * Tests * Black Co-authored-by: Ubuntu <ubuntu@ip-172-31-27-149.us-east-2.compute.internal> Co-authored-by: Ubuntu <ubuntu@ip-172-31-19-34.us-east-2.compute.internal>

* [FIX] Remove leftovers from check_correctness * remove unused numpy import

…he#7273)

…e#7280)

…checking (apache#7278) * Correct handling of call node attrs to handle non-operator calls (attrs may be undefined) * Linting fix

* Add MicroTVM support for the STM32F746 Discovery board Signed-off-by: Tom Gall <tom.gall@linaro.org> * Add reference to the discovery board in the docs Signed-off-by: Tom Gall <tom.gall@linaro.org>

* Add if pattern commit 1ee052fd494a5bdd881c242c3ea0c95cf2a613e5 Author: Masahiro Masuda <masahi129@gmail.com> Date: Sat Dec 26 22:19:17 2020 +0900 add comment commit c846a6999e9c9e48fbc019780e705a990f46cb22 Author: Masahiro Masuda <masahi129@gmail.com> Date: Sat Dec 26 21:14:20 2020 +0900 max_out_size rewrite added to the test commit 2c7c7fbd0e6563aba694e7fb6baa7bda8e4fadca Author: Masahiro Masuda <masahi129@gmail.com> Date: Sat Dec 26 20:57:55 2020 +0900 max_out_size rewrite working commit 319e930acb8162c1ec4a5d4fb71d134580a68f13 Author: Masahiro Masuda <masahi129@gmail.com> Date: Sat Dec 26 20:43:16 2020 +0900 refactor dyn strided slice pattern commit fb6917b703440748800bde624bc20efaf5798b8a Author: Masahiro Masuda <masahi129@gmail.com> Date: Sat Dec 26 11:21:33 2020 +0900 update NMS pattern following frontend change commit 255a98f1da8f300d4fe417cce3587c0d71e38ed3 Author: Masahiro Masuda <masahi129@gmail.com> Date: Thu Dec 24 05:19:31 2020 +0900 add some comment to explain the pattern commit 52cea1cc2bff533ca60acfc2416477fc8b058428 Author: Masahiro Masuda <masahi129@gmail.com> Date: Wed Dec 23 08:35:14 2020 +0900 revert tutorial change commit d3e0e0d7e2427c40067d6ad2680ec5b3f0076223 Author: Masahiro Masuda <masahi129@gmail.com> Date: Wed Dec 23 08:02:29 2020 +0900 test fixed by setting force_surpress=False commit 2fa1a574f932001be2d8f601338a342dab92f79c Author: Masahiro Masuda <masahi129@gmail.com> Date: Wed Dec 23 07:22:32 2020 +0900 fixed coord_start commit 6ba88f27dec1bdb0b0ba746c268591a59264088e Author: Masahiro Masuda <masahi129@gmail.com> Date: Wed Dec 23 06:50:46 2020 +0900 add doc commit 8d386b6a1c92ce4fe3349ff20e320199a1b5b310 Author: Masahiro Masuda <masahi129@gmail.com> Date: Wed Dec 23 05:27:26 2020 +0900 updated tutorial commit 3206b49ecfdd874e0ff8feb0fa586c4c4282f705 Author: Masahiro Masuda <masahi129@gmail.com> Date: Wed Dec 23 05:04:44 2020 +0900 update object detection test to add rewrite commit 74bebb2f4376aeb67d8c4aad395f9f2661fe6b3e Author: Masahiro Masuda <masahi129@gmail.com> Date: Wed Dec 23 05:02:15 2020 +0900 add a pattern to rewrite nms to batched nms commit f410e6dde0ed949b90312c5a7ddbb6c234f9acc1 Author: Masahiro Masuda <masahi129@gmail.com> Date: Sat Dec 26 22:20:16 2020 +0900 add comment commit f1e078b0724bd22e7be0a812055e1c7c650d94da Author: Masahiro Masuda <masahi129@gmail.com> Date: Sat Dec 26 19:54:22 2020 +0900 Add if pattern * add doc * add test * doc formatting * cpplint fix

…for Cuda & X86 (apache#7148) * [Frontend][Tensorflow] Sparse_Dense Op CSR scheduling issue resolved for both cuda & x86 * [1] Review comments handled * [2] Review comments handled * [3] Review comments handled

…#7254) Currently tutorial script 'micro_tflite.py' assumes that all boards with target STM32F746 are Nucleo boards. As a consequence once that target is selected the script automatically defaults to the Nucleo board. However, the STM32F746 is also used on Discovery Kit boards (aka disco) which are quite similar but have some differences, so Nucleo config and final image don't work on the disco boards. That commit adds a way to select a different dev board and adds comments accordingly, informing how to use the script with STM32F746 disco boards. Signed-off-by: Gustavo Romero <gustavo.romero@linaro.org>

* Upstream fix in tlc-pack/tlcpack-sphinx-addon@995178d

…7210) * made TShapeDataDependant array * add stub * dyn strided slice working * reshape also working * remove log * works on maskrcnn * lint fix * fix cpp test * remove stale pop back * add more doc * dependant -> dependent * remove redundant check * remove data_dependent_

This commit mainly introduces a byoc c-source module example to uTVM. Moreover, it carries certain modifications to the example codegen_c external module generator code to generate utvm friendly c-source. Change-Id: I09f3a42017d518dd5b6c89e3fe0a0332b80088b0

* Add fix and unit test for const autoconvert dtype. * formatting * Address review comment, casting input value to int32 * Fix failing test * Augment unit test

…uirement (apache#7294) * this test current sets a requirement to "uses_gpu", which causes it to fail in cpu-only machine * this patch changes it to be "requires_tensorcore", as per discussion on issue apache#7277

…7298)

…eld (apache#7306) [TIR][REFACTOR] ForNode update - Remove deprecated device_api. - Add ThreadBinding for_type. - Add additional annotations. More style consistency refactor to make the ForNode to be consistent with rest of the codebase. - ForType => ForKind - Add constant prefix k to enum consts per Google C style - Introduce ForKind to the python side.

… during testing (apache#7300) Co-authored-by: Josh Fromm <jwfromm@uw.edu>

* improve scatter 4d init * do not launch sorting based scatter for small input * do not use hard coded num threads * separate sort based implementation * register scatter as autotvm task * add missing import * fix strategy * add dedicated schedule and dummy flop * add test tuning script * try adding dummy knob * skip random_fill when a tuning workload is from scatter This reverts commit 1fed883. * cleanup memcpy ir * remove scatter tuning script * make sure zero init arguments * add comment on why skip random init for scatter * restore ctx sync Co-authored-by: masa <masa@pop-os.localdomain>

Added an ability to infer argument shapes if shapes are not present in TFLite files. The set of networks on which the patch was tested is internal to Arm. Any help with creating unit tests would be appreciated.

* import changes from scan branch commit cf0d4fd Author: Masahiro Masuda <masahi129@gmail.com> Date: Fri Dec 25 10:12:01 2020 +0900 get valid count test working commit eb142d3 Author: Masahiro Masuda <masahi129@gmail.com> Date: Fri Dec 25 07:22:00 2020 +0900 integrate new cumsum change commit f89684d Author: Masahiro Masuda <masahi129@gmail.com> Date: Fri Dec 25 06:56:46 2020 +0900 remove ceil_div from nms commit a2ad4de Author: Masahiro Masuda <masahi129@gmail.com> Date: Sun Dec 20 20:36:34 2020 +0900 add api for returning reduction from ex scan output commit b7f4ef7 Author: Masahiro Masuda <masahi129@gmail.com> Date: Sun Dec 20 19:49:07 2020 +0900 move ceil_div to utils commit a9a57e3 Author: Masahiro Masuda <masahi129@gmail.com> Date: Sun Dec 20 19:38:15 2020 +0900 rename prefix_scan.py to scan.py commit 03ed43f Author: Masahiro Masuda <masahi129@gmail.com> Date: Sat Dec 19 06:12:55 2020 +0900 surpress cpplint commit abceac9 Author: masa <masa@pop-os.localdomain> Date: Fri Dec 18 20:36:24 2020 +0900 support more data type commit 3e7d1f8 Author: masa <masa@pop-os.localdomain> Date: Fri Dec 18 20:09:51 2020 +0900 1d thrust scan working commit ac13b40 Author: masa <masa@pop-os.localdomain> Date: Fri Dec 18 19:49:25 2020 +0900 adding thrust scan support commit 65634e8 Author: masa <masa@pop-os.localdomain> Date: Fri Dec 18 19:01:11 2020 +0900 add thrust scan python stub commit 9876c90 Author: masa <masa@pop-os.localdomain> Date: Fri Dec 18 20:55:14 2020 +0900 introduce prefix_scan.py and move scan ir in nms.py commit 667bdd3 Author: masa <masa@pop-os.localdomain> Date: Fri Dec 18 15:06:18 2020 +0900 make the scan loop exclusive commit 480787b Author: mbrookhart <mbrookhart@octoml.ai> Date: Thu Dec 17 10:01:11 2020 -0700 Parallelize cumsum in get_valid_counts * fix for 1d scan * rename * cast to out dtype * do not run return reduction for inclusive scan * remove another ceil_div definition * adding scan test * add scheduling for scan op, fixed scan 1d test * pylint fix * add doc string * add more thrust scan test * add dynamic get valid count test, including empty size tensor * fix hard coded gpu targets for cpu only env * try retunring early if scan_size is 0 * another change for empty tensor and thrust path Co-authored-by: masa <masa@pop-os.localdomain>

Removed checks for padding in according with changes in ACL 20.11 *ACL stands for "Compute Library for the Arm® Architecture"

…6998) * [TFLite] Strided slice handlig of shrink_axis_mask improved 1. Added removal of dimensions if result is a scalar to mimic TensorFlow behaviour. E.g.: tf.strided_slice([1,2,3], [0], [1], [1], shrink_axis_mask=0) <tf.Tensor: shape=(1,), dtype=int32, numpy=array([1], dtype=int32)> tf.strided_slice([[[1,2,3],[4,5,6],[7,8,9]]], [0, 0, 0], [3, 3, 3], [1, 1, 1], shrink_axis_mask=7) <tf.Tensor: shape=(), dtype=int32, numpy=1> 2. Added extra check to assert_allclose to check shape equalities as np.testing.assert_allclose() does not distinguish between cases like: np.testing.assert_allclose(1, np.array(1)) np.testing.assert_allclose(1, np.array([1])) np.testing.assert_allclose(np.array(1), np.array([1])) * unit tests fixed

* use ex scan to write argwhere * add doc

@jwfromm

* [COMMUNITY] @jwfromm -> Committer * add areas

…er generator. (apache#7266) * [uTVM] Add TVMPlatformGenerateRandom, and use with Session nonce. * This change is preparation to support autotuning in microTVM. It also cleans up a loose end in the microTVM RPC server implementation. * Randomness is needed in two places of the CRT: 1. to initialize the Session nonce, which provides a more robust way to detect reboots and ensure that messages are not confused across them. 2. to fill input tensors when timing AutoTVM operators (once AutoTVM support lands in the next PR). * This change adds TVMPlatformGenerateRandom, a platform function for generating non-cryptographic random data, to service those needs.

* Made tensorflow IsNan actually work IsNan was added to tensorflow.rst in fa1b859 but this commit makes IsNan actually work * Added test case for tensorflow.is_nan

…he#7295) * Fix an issue with dynamic functions overwritting call arg types * fix a bug for un-annotated inputs * normalize names in TypeSolver::Unifier * fix name normalization

…pache#7286) * add files * remove import * remove os import * reorder header * fix header order cpplint * lint fix

)

* Add a relay LetPattern * fix If copy Co-authored-by: Cody Yu <comaniac0422@gmail.com> * fix If copy Co-authored-by: Cody Yu <comaniac0422@gmail.com> Co-authored-by: Cody Yu <comaniac0422@gmail.com>

* arm tuning tutorial * adjustment to get RPC working * fix lint * fix target * integrate Leandros comments * dont request remote in CI * use API from auto_scheduler, not autoTVM and updated comments * make ci-runnable * fix the formatting * address Zhaos comments * full run stats * taking Zhaos comments into consideration

…ring (apache#7317) * [AutoScheduler] Separate shapes from DAG hash and enable schedule sharing * Update CI logs * lint * fix registry * add message; fix layout rewrite mismatch * update message * support other formats

… not exist (apache#7308) * [FIX] Infer input shape in sparse_dense_padded's alter_op if one does not exist If there are multiple alter_ops in a model, the first alteration does not run type inference for the subsequent ones. In this case, we don't have the shape information, so we run the inferencer manually. * add todo

catching polymorphic type 'struct dmlc::Error' by value

* add more gradients * add documentation

…#7340)

Signed-off-by: windclarion <windclarion@gmail.com>

* In apache#5921, resource_handle was added as a parameter to TVMBackendPackedCFunc, which is the typedef for functions called by LibraryModule's function lookup. * It appears TVM_DLL_EXPORT_TYPED_FUNC was overlooked in that PR, although there don't seem to be any runtime affects known so far. However, making this definition proper to avoid any compiler warnings/debug tool problems. * See also https://discuss.tvm.apache.org/t/rfc-misra-c-changes-for-rpc-support/7098/5

* Add cumsum relay/topi op * relay tests working * add torch frontend converter * fix for importing detr * fix bad merge * begin cuda cumsum * support non innermost axis * support rank higher than 3 * making binop parameter * fix overflow issue in thrust scan * generic binop parameter working * relay test working * fixed for bool input * remove pytorch change * fix pylint * doc update * Update python/tvm/topi/cumsum.py Co-authored-by: Tristan Konolige <tristan.konolige@gmail.com> * Update tests/python/relay/test_op_level3.py Co-authored-by: Tristan Konolige <tristan.konolige@gmail.com> * add example outputs * add supported input and output dtype in thrust log * adding more loop var names * fix cpplint * fix missing check for the cuda target in nms thrust sort * parallelize cpu cumsum * making binop argument tir function * update doc for binop * doc update Co-authored-by: Tristan Konolige <tristan.konolige@gmail.com>

…ORT_PACKED_FUNC macros in packed_func.h. This is a patch PR for apache#7388. (apache#7343) Co-authored-by: JC Li <jinli@nvidia.com>

…apache#7342) If the user has a dmlc-core directory next to the tvm directory, this dmlc-core directory will be incorrectly used when compiling files with cc.py.

…c is wrapping (apache#7287) * [PRNG] Add check to PRNG to make sure that unsigned integer arithmetic is wrapping * Add threefry_test_wrapping: a manual test for wrapping unsigned arithmetic. * fix test to actually run on the target * formatting * lint

* add conversion for detr * remove explicit broadcast_to before batched matmul * use take with wrap mode * add test for transformer and negative indices * add sort and argsort * add logical_and * support masked_select * add gpu targets to masked_select test * improve sort conversion

) * [AutoScheduler] Enable schedule sharing in dispatch context * Update python/tvm/auto_scheduler/dispatcher.py

* add post nms topk to max_out_size rewrite * add argsort conversion * scatter pattern first cut * matching seems to working * dup matching fixed * add converter * conversion seems working * add reshape, use take * remove pytorch argsort converter * update test * add doc

* fix unstable compute * fix * fix * lint * sort linear equation * sort inequalities * fix * fix find * lint * fix find * lint

* Add test for array loop. * Fixed scalar issue. * Formatting. * Fix injective schedule for dynamic shapes.

… wildcard, allow grouping via dominator analysis (apache#7355)

* Various fixes to get nRF5340 working. Not yet there. * nRF5340 test runs locally. * Various fixes to get nRF5340 working. Not yet there. * nRF5340 test runs locally. * Add `nrfjprog --recover` for nRF5340DK * Cleanup. * Remove debugging code. * Revert submodule update. * Remove debugging code. * Fix comment. * Remove -keys argument. * Adding some debugging code * Fix passing west command to ZephyrFlasher. * Various fixes to get nRF5340 working. Not yet there. * nRF5340 test runs locally. * Add `nrfjprog --recover` for nRF5340DK * Cleanup. * Various fixes to get nRF5340 working. Not yet there. * nRF5340 test runs locally. * Remove debugging code. * Fix comment. * Remove -keys argument. * Fix merge.

…he#7267) * [Frontend][Tensorflow] Sparse dense matmul adjoint option added * [1] Review comments handled * [2] Review comments handled * [3] Review comments handled

When using pattern with attr of functions, such attrs mostly does not exist for op node. Therefore, hasattr check has to be done for op nodes. Change-Id: Ia313ab34be95ccc793c32fd8e5e5ef566b78685b

* [RUNTIME] Improve error messages for TypedPackedFunc - TypedPackedFunc now prints the function name when the incorrect number of arguments is passed. - TypedPackedFunc now prints the function name and which argument when an argument cannot be converted to the correct type. * check argument conversion by template deducing argument types * switch from template approach to TVMMovableArgValueWithContext * move passes back into cc files * remove error message prefixes * Remove TVM_ICHECK_TYPE_CODE. Rename name to optional_name. * revert changes to module pass for later PR * reverted too much * documentation * formatting * more docs * unify error message language. TypedPackedFunc contrustor that does not take a name * Update include/tvm/runtime/packed_func.h Co-authored-by: Junru Shao <junrushao1994@gmail.com> Co-authored-by: Junru Shao <junrushao1994@gmail.com>

* fix an error in the dynamic Full Type Relation * Add Diagnostic Errors to Broadcast Type Relations

…ed bugs. (apache#7364) * Add testing for datatypes and fix related bugs. * Fix lint issue in onnx.

* use lowercase for verilator runtime registry function * lint fix * update comment

* Port memory_alloc to c++ * remove memory python pass

* add pad_to_tensorcore & legalize for dense/bmm/conv2d * fix pad & slice * fix comments * fix comments * resolve conflict * resolve conflict * support only fp16 * add tests/python/relay/test_pass_legalize_tensorcore.py * add tests for legalize tensorcore * fix pylint * fix pylint * code format * use_gpu test only; fix conv2d_alter_op * fix tests params * revert transform fix

Co-authored-by: 袁航剑 <yuanhangjian@bytedance.com>

* fix tokenizing inf * use ParseNumber to parse inf, handle -inf * fix neg handling * fixed multi negation * refactor * use while loop * simplyfing * fix lint * simpler implementation per altan's suggestion * disable flaky test

* DynamicToStatic Refactor * fix test * add regression tests * cleanup * skip PrepareInput if the arg is already a constant * fix an issue with type inference with global functions

* [WIP][Relay][Passes] non-recursive a-normal traversals * fix clang warning * Refactor ANormal Iterative traversal into a higher order function utility with lambdas * refactor missed pass * add explict use of to lamdbas

…ction calls (apache#7330) * [FIX] Improve error messages when array/map types do not match in function calls * missed some places for renaming * Rename Mismatch to CheckAndGetMismatch. Add Check back in. Use Optional::defined. * Optional<String> -> String * formatting * move ObjectTypeChecker template specializations into where thier respective classes are defined so they will always be found correctly

* [TOPI] Einsum * Fix tuple * fix oshape * * test * * Fix lint * * Remove useless define * * Move to einsum header file * * Fix single value situation * * Fix CamelASE * * Print stride * * Fix single input bug * * fix lint * * Fix lint and add comments * * create test einsum * * Fix lint * * Fix comments

* [TFLite] Added check for dynamic range quantization Added check to prevent optimized with "dynamic range quantization" tflite files to be loaded as the optimization is not fully supported. https://www.tensorflow.org/lite/performance/post_training_quantization#dynamic_range_quantization * linter * linter * unit test fix

* Generate requirements.txt from Python spec. * add tests, collect actual requirements (first cut). * add tornado and cloudpickle * add xgboost * add xgboost version restriction * cleanup and prepare for merge * black format * add type annotations and docstrings * remove example requirements.txt * fix setup.py extras_require * use typing. classes for type annotations, python 2 compatible :) * fix python2 typing.Pattern * retrigger CI * address comaniac comments * retrigger ci

…g tasks (apache#7395) * [Bugfix][AutoScheduler] Fail to register ComputeDAG when deserialize tasks * fix test * trigger ci

* Support negative pad values * Update test_op_level2.py * Update pad.cc * Update test_op_level2.py * PR Comments * Update pad.cc * Address PR Comments * CI Error * CI Error * CI Error Co-authored-by: Ubuntu <ubuntu@ip-172-31-28-115.us-east-2.compute.internal>

…pache#7397) * Fix Bug in Bilinear Interpolation * Add NHWC Tests * clean * Fix Bug and Add Deformable Conv PyTorch for completeness * Add Tensor Utils * Remove stuff * Include vector * PR Comments * Empty Commit for CI Co-authored-by: Ubuntu <ubuntu@ip-172-31-42-251.us-east-2.compute.internal>

* [AutoScheduler] Support early_stopping per task * address comment * fix test * Update python/tvm/auto_scheduler/task_scheduler.py * Update python/tvm/auto_scheduler/task_scheduler.py * trigger ci * trigger ci

…apes. (apache#7394) * Refactor where importer to support dynamic shapes. * Add a test for dynamic where.

* Add cuda tags and unit test * Add missing space * Remove extra indent * Modify macro def position * Fix clang format * Fix clang format for set_config

* started moving things to header * directly call InvokeTVMOp * done all memory op * also refactor AllocTensor * declare Prod * remove cached func for Add, Multiply, Divide * lint fix * revert test change * remove tensor.h and declare Prod in pattern_utils.h

In case of non cpython flavour of python, the task passed to measure process should be serialized using pickle approach. The task includes workload which is a list of Tensors. The list should be serialized and deserialized as an atomic object.

Fixes incorrect number of template parameters in call to sort() Signed-off-by: Matthew Bentham <matthew.bentham@arm.com>

* support vmware_desktop provider for microTVM reference VM * update tutorial * python format * try to fix sphinx warning * fix sphinx warning * retrigger CI

* Generate JUnitXML from pytest. * address tkonolige comments

* [FIX,CMAKE] Only compile runtime files once * copy defines to tvm_runtime_objs

* add ability to optionally overide tvm shapes * add help documentation for --shapes * improve documentation * reformat test_compiler using black * Incorporate feedback from ekalda for better pytorch support and testing. * address feedback * switch input shape syntax to be more pythonic * add commentary * reformat common.py * fix lint issue * format common.py with black * torch/pytorch test hiccup * add -s to setup-pytest-env.sh for clearer error msgs Co-authored-by: Jocelyn <jocelyn@pop-os.localdomain>

* [AutoScheduler] Add sampling to dispatcher * address comment * make measurment configurable

* [ONNX] Add CumSum operator to ONNX frontend * Fix lint and add attributes to CumSum * Fix CumSum test * Add support exclusive attribute * Add support reverse attribute * Fix clang-format * Fix lint * Move reverse calculation to ONNX frontend and add exclusive to GPU * Add test for int type

* Add CPU dense weight transform * Fix format * Fix python format * Fix pylint * Minor fix * Add test * Do not need to infer layout for dense * Fix test * Rename dense_pack * Fix test * Fix lint * Fix dynamic shape dense * Fix lint * Fix autotvm task extraction test * Disable AlterOpLayout in micro_tflite.py tutorial

Clang flags were set for all file types, causing nvcc to error out.

* Dynamic Reshape * Changes * Add test cases * Add test cases * PR COmments * CI Error * EmptyCommitCIError Co-authored-by: Ubuntu <ubuntu@ip-172-31-42-251.us-east-2.compute.internal>

* convert argwhere(full(const)) to reshape(arange()) * Add IsWildcard syntatic sugar * add a simplify expression to fold full into broadcast ops * Allow constant folding of full-like ops after SimplifyExpr * fix a bug with the Attr Pattern matching * remove skip_list

…ions < 6 (apache#7372) * Fix small bug preventing TRT runtime compilation for versions < 6 * Trigger ci

* Make the TVM targets list available in Python Change-Id: I8602723fe57aaf32cee5392d4387a637115dd363 * Rename the APIs to get target kinds Change-Id: I2e6e32e025e3614a148a30a31e5a2c52fd3563cc

The type punning in the existing code is undefined behaviour in C. In particular, the existing code fails when running on Arm Cortex-M devices. On Cortex-M, accessing a uint64_t that is not 8-byte aligned generates a hard fault. Change-Id: I2aecaa220e581af7c91a8bc7886499d70e2aa6f2

…#7405) Co-authored-by: honghua.cao <honghua.cao@streamcomputing.com>

* [AutoScheduler] Fix distill record * update comments

…he#7447) * fix missing return in scatter_nd cuda strategy * add Relay test for scatter_nd, fix documentation

* Build microTVM using standalone_crt in build tree. * black format * pylint * try stashing entire standalone_crt in hopes it will not upset jenkins * Put standalone_crt in correct Jenkinsfile stash bundle * include build prefix * switch to python script for expanding globs * revert attempt to use globs in pack_libs, switch to building standalone_crt * properly revert pack_lib changes * fix typo * retrigger CI * revert pyproject.toml * update Jenkinsfile approach to use task_ci_setup.sh

* Construct static Ops if inputs are Constant * Expose FoldConstant as a function in addition to the pass * refactor onnx importer to do more static imports by constant folding fix pylint * fix test regressions * fix style, two bugs * pipe freeze_params through sub_graphs when importing loops and control flow

* [VM] Move param bind to OptimizeModule * add test to verify the number of free vars after opt * remove const from OptimizeModule

* Add support for MXNet GroupNorm * Fix python lint * Fix lint

Add the -mcpu information to complete the picture. Signed-off-by: Vincent ABRIOU <vincent.abriou@st.com>

* [FRONTEND][TFLITE] get input tensor information from graph * remove bare-except * fix lint * delete empty line * comment change * move some of the tflite frontend code from tvmc to tflite.py * update shape and dtype when user provided them * remove unused var. pass user provided shape_dict * remove duplicate code

…pache#7449) * [µTVM] Print .elf statistics for a model runtime built with Zephyr Currently there isn't any statistics about the used resources by a model runtime built with Zephyr, making it difficult to have any idea about, for instance, the amount of memory taken by the operations necessary to run the model. Since Zephyr's SDK already exposes the statistics about various memory regions on linking by passing '--print-memory-usage' to the linker, it's possible to use it to have an idea about the amount of memory used by the model and how much memory is left on the device. That commit adds a simple method to extract the memory region information out of the build output and then uses it to show memory usage statistics for various memory regions when Zephyr finishes building the image to be flashed to the target device. Signed-off-by: Gustavo Romero <gustavo.romero@linaro.org> * v2: Fixes accordingly to Andrew review - Catch StopIteration in case of a weird output or no additional lines after the last memory region - Use of _LOG.info() instead of plain print() for better control over the output by the main script - Set log level in micro_tflite.py script as an example on how to get the new memory usage statistics and also because currently that's the main script used to test microTVM + Zephyr's SDK - Improve statistics header Signed-off-by: Gustavo Romero <gustavo.romero@linaro.org> * Fix build It seems build system is using Python < 3.7, so 'text' argument is not present as an alias for 'universal_newlines'. To satisfy it use old 'universal_newlines' argument which is available prior to Python 3.7. * Fix build Avoid exception anti-pattern when catching StopIteration * Retrigger CI

* Add frontend code and tests * Add Frontend Code Co-authored-by: Ubuntu <ubuntu@ip-172-31-42-251.us-east-2.compute.internal>

* This PR makes it possible to type `docker/bash.sh ci_cpu tests/scripts/task_config_build_cpu.sh` and the same version of ci_cpu as is used in Jenkins will be used to run the command.

* new experiment * save * refactor * refactor library * add profiler * refactor * refactor * add docs * update comment * add deallocator

* ROI align with max on cpu passes * onnx test file was not running gpu testsgit status! * all passing * fix lint * lint again * lint * lint * typo * remove import * fix import * add inf, -inf to hybridscript and respond to comments * shorten code * make atol lower

* enable rocm thrust, confrimed to work on sort and scan * add rocm argsort strategy * Abort if CXX is not hipcc * add more strategy * add missing import * fix lint * show supported data type in err msg * try remove rocthrust * add missing include for rocthrust * more minor change Co-authored-by: Masahiro Masuda <masahi@129@gmail.com>

* Initial Commit * Fix formats * Remove comments * Black * THreeops * Add Frontend Code * Add Default Value to feed dict * Add Frontend Code * New test Cases and new code to handle them * Add Python Implementation' ' * Remove stuff * Remove unused imports * Pylint * Pylint * PyLint Shape Func * Make tests cpu only * Add unsorted tests * Add frontend code * Row Major Sorting Only Test * Handle Dynamic Shapes * Add dynamic input shapes * Dynamic Shape Tests * Add documentation * Dtypes * PR Comments * Added comments and changed naming * Add comments * Comments to Shape Func * Documentation * PR Changes * PR Comments * Resolve input and output dtype compat Co-authored-by: Ubuntu <ubuntu@ip-172-31-42-251.us-east-2.compute.internal>

…e pattern or constructor with no args (apache#7459) * [match_exhaustion] Fix cartesian product to handle empty tuple patterns or constructors with no args * Test cases do not actually exhibit the fixed bug * Mistake in comment

* Enable JUnit parsing for Python tests * retrigger CI * prefix junit results with FFI type * remove - in junit prefix

Note: 'ETHOSN_VARIANT_CONFIG' must be set to test against Ethos-N78 and this adds support for one configuration of Ethos-N78 in TVM.

* begin nhwc roi align * integrate mode change from upstream * adding test * support nhwc shape func * update strategy * refactoring test * refactor test * refactoring * fix lint * update relay op tests

apache#7454) * Fast path for single thread run to allow app level threading * add sync counter to avoid error in one of tests

* [Torch] Add index_put operator * Skip test_frontends.py::test_load_model__pth

…apache#7467) * Fix off-by-one in BiasAddRel, use new reporting * No need to mark xfail if the exception is caught * lint

Change-Id: I6a6e78080f36e4e3e1689e03ea48e759fcd8e466

…#7304) * Extend --target syntax to cover multiple targets for compilation and tuning * Add a new composite_target module to implement custom codegen passes into TVMC * Provide implementation to integrate TVMC, to target Arm Ethos-N NPU and Compute Library for the Arm Architecture (ACL) Change-Id: Iaee53fe22f0c14eb4e4c8ec47e72bade0c5e32cc

) * Ignore some TF2.0 attributes * Support explicit padding for conv2d, max_pool, conv3d * Remove conv3d explicit padding test since TF API doesn't allow it

StmtFunctor applies context dependent copy on write, which requires check over all the dependency chain. Such function is better suited as a special implementation to avoid misuse. This PR refactors the code to specialize the function. Co-authored-by: Bohan Hou <32121147+spectrometerHBH@users.noreply.github.com>

) * add contrib/thrust.py * update cuda strategy * remove is_thrust_available, update nms, scan, sort and tests * remove unused import * trigger CI * update * add note on how to enable thrust in ssd tutorial * add warning * Revert "update" This reverts commit c1629b3. Co-authored-by: masa <masa@pop-os.localdomain>

Co-authored-by: mshr-h <mshr-h@users.noreply.github.com>

* Add target host field in Target * Add host as a config field to target * Add target host support for Python api * Add unit tests * Adjust format for cpplint * Remove unnecessary after in Python file * Remove redundancy and add param description * Fix format issue * Fix param description * Add unit test for duplicate target hosts

…7347) * [RELAY][Parser] Optimize relay parser to restore attrs for non-Operator calls * To avoid too much modification to the native class, only print out the attrs type key of non-Operator Call in relay printer. Then reconstruct the attrs object after parsing this attrs type key value in Relay parser. * fix lint * fix ci * add test case

* Make onnx gemm tensor C optional * fix codestyle * add tests * fix codestyle

…7398) * Create C-runtime-style metadata module for llvm builds. * maybe address manupa's comment * lint * actually address manupa comments * comment and rename * git-clang-format * pylint * cpp warning * try to fix apps/bundle_deploy * black format * build correct file * Use save() for C++-runtime targeted artifacts. * fix build_module LLVM metadata module conditions * fix test comment * black format * further restrict CRT MetadataModule creation * Fix test_link_params * black format and address zhiics comments * fix test_link_params, i think?

…ache#7481) * Fix stack overflow when partially-__init__ Node raises exception. * If a Node subclass raises an exception and ctypes is in use before __init_handle_by_constructor__ is called (or self.handle is otherwise set), a Python stack overflow could result. This is because the unset handle slot causes self.handle accesses to fallback on the getattr(self, 'handle') method, invoking NodeGetAttr. * Then I believe this causes an infinite loop. * The fix is to make Node.__getattr__ raise AttributeError for all attributes in __slots__, then make __del__ tolerant to missing self.handle. * I don't believe cython is affected because it implements a descriptor to access its underlying chandle and that shouldn't be unset. * black format * actually use handle instead of self.handle

…apache#7509) * Fix onnx gemm with non parameter C. * Add gemm tests for C. * Fix formatting.

* fix type inference for conv2d * fix

* Renames the "acl" composite target to point to the specific library it represents

…TF 2.x (apache#7502) * add TF2.x raw_ops.all axis range support * apply linting * fix range() func input

…ng & update docs and docker (apache#7350) * Fix bug in vitis ai codegen out tensor names matching & update docs & update docker * Update vitis_ai.rst * Move gpg-agent package installation to vitis ai core script * Refactor install_vitis_ai_core script * Update docs/deploy/vitis_ai.rst Co-authored-by: Cody Yu <comaniac0422@gmail.com> * Update docs/deploy/vitis_ai.rst Co-authored-by: Cody Yu <comaniac0422@gmail.com> * Update vitis-ai docs pynq/edge setup & adjustements for comments * Update python/tvm/contrib/target/vitis_ai.py Co-authored-by: Cody Yu <comaniac0422@gmail.com> * Reorg Vitis AI dockerfile to make sure gpg-agent is installed before llvm Co-authored-by: Jorn Tuyls <jornt.tuyls@gmail.com> Co-authored-by: Cody Yu <comaniac0422@gmail.com>

* [Frontend] TF V2 sparse.todense() test added * [1] Review comments handled

* Introduce code_loader to AutoTVM. * Prepares for autotuning with microTVM, and provides extension hook for VTA. * add vta hook * git-black * pylint * Add missing import * Fix import problem * add missing import * rename code_loader to module_loader * rename remote_kw to remote_kwargs * black format

* simplyfing * improved fast path for slice * update rewrite pattern for maskrcnn

… have static dim (apache#7487) * enforce static dim for non-concat axis * assign any when all dims are dyn * add missing case * simplify * add test * only enforce static dim constraint if concat output is dynamic * more update to concat type rel * update tests * fixed compile warning

* Initial commit of the unique operator Add unit tests for unique operator * Add tensorflow unique op * Refactor unique to use sort-based algorithm * Change relay.unique test to run only on cpu * Change topi.unique test to run only on cpu * Change range to parallel for parallelizable loops * Add return_counts option for relay.unique and topi.unique, add pytorch frontend * Fix pylint * Patch pytorch frontend * Initial support of topi.cuda.unique * Refactor to use ir_builder directly * Modularize adjacent difference * Refactor to simplify * Fix typo * Combine _unique and _unique_with_counts * Reuse indices_ptr to remove arange_ptr Co-authored-by: Yanming Wang <yanmwang@amazon.com>

* Convert strides and pool_size to int * Make helper function, add test * Fix lint

* SparseReshape Inital Code * Done * Format * Add empty tests * Formatting * SanityCheck * formatting documentation * Documentation * Only Enable CPU * Add support for CUDA * Stuff * Add Dynamic Support * Parallelize GPU Impl * Documentation * Documentation * Import * Import * Remove unnecessary code * PR Comments * Schedules * Tests * Dtypes * Black * Parallelize CPU * CI error Co-authored-by: Ubuntu <ubuntu@ip-172-31-42-251.us-east-2.compute.internal>

* Make topi more permissive * Remove testing stuff * lint * Downsampling tests

* fix datatype on Reciprocal op * clean up test case

* [CI] Move ci-cpu to use llvm-11 * Fix the testcase of x86 codegen by relax the register names.

* Add create_local_debug_runtime to micro exports. * retrigger CI

…pache#7512) * Don't run non-tvm_op GraphRuntime nodes in Debug Runtime over RPC. * These are filtered out in SetupOpExecs for normal debug runtime operation. * retrigger CI * retrigger CI * address tkonolige comment

Co-authored-by: Yanming Wang <yanmwang@amazon.com>

Co-authored-by: masa <masa@pop-os.localdomain>

Co-authored-by: Bohan Hou <32121147+spectrometerHBH@users.noreply.github.com> Co-authored-by: Junru Shao <junrushao1994@gmail.com> Co-authored-by: Tianqi Chen <tqchen@users.noreply.github.com> Co-authored-by: Ruihang Lai <lairuihangdongdong@qq.com> Co-authored-by: Hongyi Jin <3231950289@qq.com> Co-authored-by: Wuwei Lin <wuwei@apache.org> Co-authored-by: Bohan Hou <32121147+spectrometerHBH@users.noreply.github.com> Co-authored-by: Junru Shao <junrushao1994@gmail.com> Co-authored-by: Tianqi Chen <tqchen@users.noreply.github.com> Co-authored-by: Ruihang Lai <lairuihangdongdong@qq.com> Co-authored-by: Hongyi Jin <3231950289@qq.com> Co-authored-by: Wuwei Lin <wuwei@apache.org>

) * fix type inference for conv2d * fix * adding the autoscheduler layout rewrite pass to VM compiler passes * revert edits applied in other PR * minor fix * fix * formatting fix * lint

* fuse constant padding into conv kernels * change the kernel to support other layouts * add channel-last test * add a comment about bailing early

* fix: cuda codegen vectorize cast * style: fix python coding style * fix: missing break * refactor: directly split by factor Co-authored-by: jiangchengquan <jiangchengquan@bytedance.com>

…he#7549) * Fix converting torch slice op with dynamic slice length * use isinstance Co-authored-by: masa <masa@pop-os.localdomain>

* basic pass profiler prototype * allow enable/disable of pass profiling * lint * add example pass profiler usage as test * render pass profiles to String instead of stdout

* add while node * update visitors * binary search lowering works * llvm codegen working * cuda codegen working * nms updated to use while loop * add missing upper bound check too * add mandelbrot test * add gpu mandel commit ee2363b Author: Masahiro Masuda <masahi129@gmail.com> Date: Fri Jan 29 11:44:02 2021 +0900 enable extern lib offload for nvptx * rename test * run black * add doc * add collatz test * add while + vectorize test * simplify bin search * Add special case visit method to storage_access.cc * disallow while loop inside vectorized loop * disallow trivial condition since we do not have break * error out in CoprocSync for now * error out LiftAttrScope for now * add placeholder to inject_vpthread * refactor to use MakeAttach * handle WhileNode in InplaceOpVerifier * error out in InjectVirtualThread * try handle WhileNode in StoragePlanRewriter * remove WhileNode visitor from storage rewrite * add while loop storage rewrite test * update tests * move test_vectorize_while_fail to test_tir_transform_vectorize.py

…e#7558) * [RELAY] Modify some passes to not stack overflow on many lets. Passes modified: - inline primitives - dead code - lambda lift * one fix * small fix * .at -> [] * fix

…scope (apache#7497) * [Tensorize] Support conds depend on outer loop vars inside tensorize scope * Reformat

* Update Vitis AI CI PyXIR version to v0.1.6 * Add --depth 1 to PyXIR clone command

* Add SPIR-V lowering for WhileNode * test vulkan in while loop tests

…che#7432)

…tive index being out of bounds (apache#7554)

* [RUNTIME] Move Map into runtime This allows us to use Map to store parameters needed at runtime. * node.{Array|Map} -> runtime.{Array|Map} * missed some renames

* [AutoScheduler] Query in task extraction * trigger ci

…apache#7579)

) - Updated ethosn relay backend to support 20.11 api changes. - Removed legacy support for 20.05. - Added a mechanism to specify the ethosn driver stack version.

* Rewrite the Rust Module API and change some imports causing crashes. This commit also updates the docs to remove outdated information. * Renable Python test and remove warnings * Python test still flaky * Fix broken module test * Fix broken test * Reset test file

…add dynamic bug (apache#7562) * Add segment sum Op * Remove unnecessary * Documentation * Black * Add GPU * Uncomment * Add documentation * Add dynamic tests * Add TF Op * Add Sparse Segment Sum * Add test coverage * PR Comments * Int64 tests * Add SparseSegmentSqrtN * Add SparseSegmentSqrtNOp * Deduplicate code * Add SparseSegmentMean * Parametrize Tests * Remove * Modularize * Black * Modularize Code * Pylint * PR Comments * Add scatter add tests * Remove Test Co-authored-by: Ubuntu <ubuntu@ip-172-31-42-251.us-east-2.compute.internal>

…pache#7581) * Prevent TRT runtime crash for duplicate inputs and outputs * Add empty subgraph unit test

* Use correct default value of False for is_ascend * Add unit test for default topk is_ascend value

…conv when groups>1 (apache#7595) * trt num_outputs * asdf * fix lint Co-authored-by: Leyuan Wang <leyuan.wang@bytedance.com>

* Fix negative axis in gather * Clang Format * Black * Empty Commit Co-authored-by: Ubuntu <ubuntu@ip-172-31-42-251.us-east-2.compute.internal>

Co-authored-by: Wuwei Lin <wuwei@apache.org>

* Fix autotuning, broken in apache#7337 * retrigger CI, because I don't understand how it passed

…e#7588) * init * fix * fix

apache#7539 Co-authored-by: guoweijun <guoweijun@baidu.com>

…pache#7313) * Add sparse dense tuning tutorial * Add sparse input fusion * Update the dag to support output fusion * Update * Add task input to search_task * Update * Add search_inputs to measure * Lint fix * Lint fix * Update * Update * Update * Update * Add file save load support * Update * Update * Update * Remove add_task_inputs API * Update * Update * Update * Lint fix * Lint fix * Lint fix * Lint fix * Update * Add example ci_log * Update * retrigger ci * Update * Update * Update * Lint fix * Lint fix * Lint fix

…ache#7603) * Move SimplifyConvPad to a new pass and don't enable it by default * rename pass * move files * fix lint * adjust test tolerance

…ecutor (apache#7604) * properly return and unflatten outputs from GraphExecutor * lint * cleaner approach, not sure what I was thinking before * remove unused import * forgot copyto cpu * make solution even cleaner using iterator

* [Torch] support hardsigmoid * qhswish first impl * add qhardsigmoid but the result is not correct * add qmv3 to test * comment fix

…pache#7602) * [TE] Fix bug in AutoInlineElemWise and implement AutoInlineBroadcast * [TE] Add AutoInlineBroadcast API to schedule_pass.h

* add ShapeFunc for tanh * _schedule_dense_small_batch turn autotvm off when dense's inner dim is unknown * fix CI pylint

* [Relay] Fix relay op strategy for cuda dense int8 * Remove uint8 && Add autotvm task extraction test for relay graph that contains dense op (int8 * int8 -> int32) * Reformat the code of test case

* [Relay] add ShapeFunc for one_hot op * fix pylint * add test for shapefunc of one_hot op

…ache#7617)

…#7607) * sort started to working * static size sort seems to be working * test sort on vulkan * add nvptx to sort test too

Fix TVM compilation with USE_LLVM=OFF #194

Fix TVM compilation with USE_LLVM=OFF #194

Commits on Jan 13, 2021

Commits on Jan 14, 2021

Commits on Jan 15, 2021

Commits on Jan 16, 2021

Commits on Jan 18, 2021

Commits on Jan 19, 2021

Commits on Jan 20, 2021

Commits on Jan 21, 2021

Commits on Jan 22, 2021

Commits on Jan 23, 2021

Commits on Jan 25, 2021

Commits on Jan 26, 2021

Commits on Jan 27, 2021

Commits on Jan 28, 2021

Commits on Jan 29, 2021

Commits on Jan 31, 2021

Commits on Feb 1, 2021

Commits on Feb 2, 2021

Commits on Feb 3, 2021

Commits on Feb 4, 2021

Commits on Feb 5, 2021

Commits on Feb 6, 2021

Commits on Feb 7, 2021

Commits on Feb 8, 2021

Commits on Feb 9, 2021

Commits on Feb 10, 2021

Commits on Feb 11, 2021

Commits on Feb 12, 2021

Commits on Feb 13, 2021

Commits on Feb 14, 2021

Commits on Feb 15, 2021

Commits on Feb 16, 2021

Commits on Feb 17, 2021

Commits on Feb 18, 2021

Commits on Feb 19, 2021

Commits on Feb 20, 2021

Commits on Feb 21, 2021

Commits on Feb 22, 2021

Commits on Feb 23, 2021

Commits on Feb 24, 2021

Commits on Feb 25, 2021

Commits on Feb 26, 2021

Commits on Feb 27, 2021

Commits on Feb 28, 2021

Commits on Mar 1, 2021

Commits on Mar 2, 2021

Commits on Mar 3, 2021

Commits on Mar 4, 2021

Commits on Mar 5, 2021

Commits on Mar 6, 2021

Commits on Mar 8, 2021

Commits on Mar 9, 2021