platform/upstream/tvm.git
4 years agoAdd rocblas_sgemm_strided_batched impl. (#6579)
Chris Sullivan [Tue, 29 Sep 2020 09:59:54 +0000 (02:59 -0700)]
Add rocblas_sgemm_strided_batched impl. (#6579)

4 years ago[COMMUNITY] vegaluisjose -> Committer (#6582)
Yizhi Liu [Tue, 29 Sep 2020 08:05:05 +0000 (01:05 -0700)]
[COMMUNITY] vegaluisjose -> Committer (#6582)

4 years ago[tvmc] unify all logs on a single logger 'TVMC' (#6577)
Leandro Nunes [Tue, 29 Sep 2020 05:15:36 +0000 (06:15 +0100)]
[tvmc] unify all logs on a single logger 'TVMC' (#6577)

4 years ago[Doc] Update release document (#6573)
Yizhi Liu [Mon, 28 Sep 2020 17:41:09 +0000 (10:41 -0700)]
[Doc] Update release document (#6573)

4 years agoSupport mxnet dot and LogisticRegressionOutput (#6542)
Trevor Morris [Mon, 28 Sep 2020 17:35:01 +0000 (10:35 -0700)]
Support mxnet dot and LogisticRegressionOutput (#6542)

4 years ago[tvmc] Fix command line argument variable name (#6574)
Leandro Nunes [Mon, 28 Sep 2020 16:50:00 +0000 (17:50 +0100)]
[tvmc] Fix command line argument variable name (#6574)

* This is a variable that we changed names during code review,
   and it was incorrectly left behind with the previous name.

4 years agoFix android runtime error (#6575)
eric [Mon, 28 Sep 2020 15:26:56 +0000 (00:26 +0900)]
Fix android runtime error (#6575)

codes for USE_RANDOM in tvm4j was missing.

4 years agoFix missing te in the code example (#6569)
isong [Mon, 28 Sep 2020 04:24:10 +0000 (21:24 -0700)]
Fix missing te in the code example (#6569)

- caused `AttributeError: module ‘tvm’ has no attribute ‘thread_axis’
  error
- Fix baesd on https://discuss.tvm.apache.org/t/attributeerror-module-tvm-has-no-attribute-thread-axis/6606

4 years ago[TEST] Temporary disable test_mutate_parallel (#6572)
Tianqi Chen [Sun, 27 Sep 2020 22:30:25 +0000 (15:30 -0700)]
[TEST] Temporary disable test_mutate_parallel (#6572)

* [TEST] Temporary disable test_mutate_parallel

* Use skip

4 years ago[AutoScheduler] Improve the rule of mutating parallel granularity (#6568)
Lianmin Zheng [Sun, 27 Sep 2020 09:18:10 +0000 (02:18 -0700)]
[AutoScheduler] Improve the rule of mutating parallel granularity (#6568)

* fix mutate parallel

* fix comments

* fix lint

* fix tutorials

* update

* fix tests

* address comments

4 years ago[BYODT] fix CMAKE flag name + update documentation (#6567)
Andrew Liu [Sun, 27 Sep 2020 09:17:39 +0000 (02:17 -0700)]
[BYODT] fix CMAKE flag name + update documentation (#6567)

* documentation fixes + better name

* fix up comments

4 years ago[APP] Fix misprint in demo.cc during initializing of picture tensor data (#6566)
dlexplorer [Sun, 27 Sep 2020 00:01:01 +0000 (03:01 +0300)]
[APP] Fix misprint in demo.cc during initializing of picture tensor data (#6566)

4 years agoBring Your Own Datatypes (#5812)
Gus Smith [Sat, 26 Sep 2020 22:49:20 +0000 (15:49 -0700)]
Bring Your Own Datatypes (#5812)

* Add ChangeDatatype pass and unittest

* [WIP] Jared's work on Fri

This was work that Jared did on my computer, trying to get Inception v3 running.

* Fix simplify inference to work over different data types.

* Formatting

* Copy setup code from other test file

* Logging in Relay

* Remove duplicate TVM_DLL

* Add Sub, Mul, Div, Max to bfloat lib

* Fix previous broken rebased commit

* Remove line

* Add LowerCustomDatatypes to build passes

* Upcast ints to custom datatypes too, as well as to floats

* Add and use convert_ndarray

* Lower Call

* Relay: create constant scalars of custom dtypes

We use the same method we use in TVM: store the value in a double.

* Custom datatype formatting in Relay

* Update unittests

* Add simpler example that's not working yet

* Add Python unittests to Makefile

* Fix bug

* Fix function name in GetPackedFunc call

* convert_ndarray makes its own executor

* Add simple test case

* Move setup() calls

* Use convert_ndarray

* Change import to make it more specific

* Fix another Registry::Get call

* Allow users to register minimum functions for custom datatypes

This commit allows users to register global functions named
`tvm.datatype.min.<type name>` which take the number of bits in the custom type
and return the corresponding minimum value (as a double).

A similar commit will need to be created for max, whenever that ends up being
needed!

* Remove check for float

* Add test

* Fix inception test

* Add MobileNet

* Lower custom datatypes before intrinsics

* Add exp and sqrt bfloat functions

* [buggy commit] Lower intrinsics like sqrt, exp

This commit has bugs in it, I'm fairly certain.

* Formatting

* Fix bug

* Add lowering for new ops in test

* Add int to bfloat

* Remove print

* Add all tests

* Correct image size

* Add TODO

* Add "notbfloat" type

This type is for testing purposes. It just stores a float in a uint32. It was
used to confirm the fact that my bfloat "implementation" is very numerically
unstable and was causing issues when running the model.

* Convert arguments

Not sure how necessary this actually is.

* Rewrite custom datatype constants in Relay

* Add test_ops

* Print constants in Relay

* Use topi.testing

* Test conv2d

* Add test_model

* Comment out model tests

* Register notbfloat

This could be unregistered at some point later

* Add commented code

Remove later

* Add posit tests

* test_ops_same_function

* [temporary] move incomplete commit to macbook

* Add more to tests

* Formatting

* Uncomment add

* Remove bad tests

* Change comments

* Change function name and docstring

* Change main function

* Restructure tests

* Fix visibility of posit functions

* YAPF

* Switching keywords around to resolve build errors on some systems

* Improve test by running smaller mobilenet

* Add test_cast

* Change datatype name; add simple test

* Rename to posit32

* Merge 3 posit types into one file

* Add a nop type

* Remove bfloat

* Refactor test comments

* Refactor conv2d test

* Add optional tolerance arguments

* Add posit8 and posit16

* Add comment about posit8

* Whoops -- actually add noptype to CMakeLists

* Add rtol, atol to run_workload

* Add noptype to tests

* Run noptype over other models, too

* Pass correct arguments to calls

* Fix line length errors

* Raise tolerances (again) to avoid flaky test

* fix style

* add test for tanh, log, sigmoid

* Remove references to bfloat, notbfloat

* Change comments

* Remove old test file

* fix min func

* refactoring unit test file

* use posits es2

* cleanup

* comment

* coment if_then_else

* support different bit widths

* use random seed to create stable tests

* update documentation

* removed nop-type and code consistency

* add batchnorm test

* rebase and update

* fix tests and format

* pylint

* change order of include

* include order

* fix style

* remove posit c linkage

* update universal

* fix style

* fix test

* fix overflow error with minfunc and posits

* style

* use change_dtype to convert params

* update universal

* fix fatal error

* fix constant repr

* minor update to posites2

* update universal

* fix rst

* fix invalid import and sqrt

* update universal

* comments

* comments and expand testing

* increase atol/rtol for custom[posites2]32

* Re-add newline

* Remove comment

* Remove opt level and comment

* Change docstring

* Add TODO

* Add file header and newline

* Update docstring

* Update file docstring

* Update docstrings

* Delete todos

* create_min_lower_func

* add better debugging message

* docs

* add BYODT tutorial

* add todo

* Reformat some of tutorial to RST, plus code fixes

* tutorial notebook runs now

* fix hyperlink

* rebase

* add to tutorial

* fix mobilenet model

* add skip tag

* black lint

* add compiler flag and add dummy float

* myfloat and posites2 test

* remove universal

* lint

* lint

* add setup

* build with USE_POSIT for CI/CD

* fix posit cmake

* add cd /

* undo docker changes

* change tutorial to use myfloat

* move files

* lint

* fix

* remove filter

* fix lint

* fix suggestions

Co-authored-by: Jared Roesch <roeschinc@gmail.com>
Co-authored-by: Andrew Liu <andrewlliu@gmail.com>
4 years agoRename tvm.hybrid.script to tvm.script. (#6522)
Tristan Konolige [Sat, 26 Sep 2020 22:32:08 +0000 (15:32 -0700)]
Rename tvm.hybrid.script to tvm.script. (#6522)

4 years agoMake CMakefile/config.cmake/install_tvm consistent (#6562)
Cody Yu [Sat, 26 Sep 2020 16:22:16 +0000 (09:22 -0700)]
Make CMakefile/config.cmake/install_tvm consistent (#6562)

4 years agoZhi's key for ASF release (#6554)
Zhi [Sat, 26 Sep 2020 03:29:40 +0000 (20:29 -0700)]
Zhi's key for ASF release (#6554)

4 years agoAdd proper cmake PATHS when multiple NAMES. (#6558)
Balint Cristian [Fri, 25 Sep 2020 16:37:20 +0000 (19:37 +0300)]
Add proper cmake PATHS when multiple NAMES. (#6558)

4 years ago[Ansor][FLAKY] Bug fix for compute at mutation error (#6557)
Chenfan [Fri, 25 Sep 2020 16:36:43 +0000 (00:36 +0800)]
[Ansor][FLAKY] Bug fix for compute at mutation error (#6557)

4 years agoMake missing desired layout non-fatal (#6553)
Trevor Morris [Fri, 25 Sep 2020 15:39:08 +0000 (08:39 -0700)]
Make missing desired layout non-fatal (#6553)

4 years ago[TIR] Fix rewrite_simplify tir::builtin::shift_left (#6555)
lixiaoquan [Fri, 25 Sep 2020 14:37:56 +0000 (22:37 +0800)]
[TIR] Fix rewrite_simplify tir::builtin::shift_left (#6555)

4 years agoFeat(frontend-pytorch): Add input types argument and Support cast to float16. (#6546)
Liu Qi [Fri, 25 Sep 2020 14:04:30 +0000 (22:04 +0800)]
Feat(frontend-pytorch): Add input types argument and Support cast to float16. (#6546)

1. Add input types argument for converting TorchScript file.
2. Support casting float32 to float16 when converting to operation.

4 years ago[COMMUNITY] Add Ziheng's key for ASF release (#6552)
ziheng [Fri, 25 Sep 2020 09:01:42 +0000 (02:01 -0700)]
[COMMUNITY] Add Ziheng's key for ASF release (#6552)

4 years agoGeneralize the use of booleans to support all cmake boolean values. (#6515)
Robert Kimball [Fri, 25 Sep 2020 05:08:12 +0000 (22:08 -0700)]
Generalize the use of booleans to support all cmake boolean values. (#6515)

* Generalize the use of booleans to support all cmake boolean valuues.

* Update to use a more simple method to detect if variable is a false value

* Fix some errors

* Debug CI issue

* Fix logic error in cmake changes

* Use new IS_TRUE_PATTERN to make intent clear

4 years ago[Relay/TOPI] Added dilation_value attribute to dilate operator. (#6550)
Rishabh Jain [Fri, 25 Sep 2020 04:06:15 +0000 (09:36 +0530)]
[Relay/TOPI] Added dilation_value attribute to dilate operator. (#6550)

* Added dilation_value attribute to dilate operator of Relay/TOPI.
  (Enables custom value for dilation, instead of always 0)
* Added tests for dilation_value of dilate operator in Relay and TOPI.

4 years ago[Frontend][Onnx] Added broadcasting to prelu alpha. (#6549)
Josh Fromm [Fri, 25 Sep 2020 00:47:36 +0000 (17:47 -0700)]
[Frontend][Onnx] Added broadcasting to prelu alpha. (#6549)

4 years agoupdate webgpu api (#6547)
WenheLI [Fri, 25 Sep 2020 00:46:43 +0000 (19:46 -0500)]
update webgpu api (#6547)

4 years ago[Rust] Allow convert Context to ArgValue (#6544)
韩朴宇 [Thu, 24 Sep 2020 15:50:49 +0000 (23:50 +0800)]
[Rust] Allow convert Context to ArgValue (#6544)

4 years agoAdd alternate cublaslt library name. CUDA 11.0 uses cublasLt. (#6541)
Robert Kimball [Thu, 24 Sep 2020 15:50:08 +0000 (08:50 -0700)]
Add alternate cublaslt library name. CUDA 11.0 uses cublasLt. (#6541)

4 years ago[Relay/TOPI] Added 'offsets' and 'alignment' attributes to MATRIX_SET_DIAG. (#6429)
Rishabh Jain [Thu, 24 Sep 2020 15:43:28 +0000 (21:13 +0530)]
[Relay/TOPI] Added 'offsets' and 'alignment' attributes to MATRIX_SET_DIAG. (#6429)

* [Relay/TOPI] Added 'offsets' and 'alignment' attributes to MATRIX_SET_DIAG.

* Added support for 'offsets' and 'alignment' attributes of MATRIX_SET_DIAG.
  (Similar to MATRIX_SET_DIAG V3 of TF)
* Added tests for 'offsets' and 'alignment' attributes of MATRIX_SET_DIAG.

* Changes by black.

* * Added offset check in Relay.
* Minor changes.

* Added more tests and some minor documentation changes.

4 years agotvmc: solve a linting error on onnx command line driver frontend (#6536)
Leandro Nunes [Thu, 24 Sep 2020 06:48:10 +0000 (07:48 +0100)]
tvmc: solve a linting error on onnx command line driver frontend (#6536)

* Updates onnx load function from "load" (a compatibility attribute)
   to "load_model" (the actual function)

 * Add a pylint command, that we don't see currently on upstream CI,
   but it reproduces when linting it locally.

4 years agoRemove unused pylint 1.9.4 from docker installation script (#6538)
Leandro Nunes [Wed, 23 Sep 2020 20:02:56 +0000 (21:02 +0100)]
Remove unused pylint 1.9.4 from docker installation script (#6538)

4 years ago[Relay] Show yolo detection result in text. (#6367)
Hua Jiang [Wed, 23 Sep 2020 16:54:05 +0000 (09:54 -0700)]
[Relay] Show yolo detection result in text. (#6367)

* [Relay] Show yolo detection result in text.

Issue:
Current yolo detection only provide a image drawing solution
to output detection result, but for console user, such result
would not available for a view.

Solution:
Here we add a text show function to show detection result in
text format for testing in console scenario.

* rebase upstream and merge code change.

* address review comments.

4 years ago[BYOC][ACL] Add maximum support for float32 (#6506)
Giuseppe Rossini [Wed, 23 Sep 2020 16:42:24 +0000 (17:42 +0100)]
[BYOC][ACL] Add maximum support for float32 (#6506)

* ACL integration: add maximum support for float32.

* Added the code generation flow in arm_compute_lib.py
* Added the runtime calls in acl_runtime.cc

Change-Id: I69c5522f05a46c1dd235da5d57fe499134de0425

* Add maximum to the list of supported functions

Change-Id: Ia49087756be4c3ac92a3dc76fe03fb00de468f8d

4 years ago[Ansor] Parallel the InitPopulation (#6529)
Chenfan [Wed, 23 Sep 2020 15:37:42 +0000 (23:37 +0800)]
[Ansor] Parallel the InitPopulation (#6529)

4 years ago[FIX] fix the python script for building resnet (#6526) (#6527)
Shuwei Liu [Wed, 23 Sep 2020 15:20:23 +0000 (23:20 +0800)]
[FIX] fix the python script for building resnet (#6526) (#6527)

4 years ago[TUTORIAL] Fix Some Failed Tutorials of The Issue #6453 (#6534)
Qiang Zhang [Wed, 23 Sep 2020 15:18:40 +0000 (23:18 +0800)]
[TUTORIAL] Fix Some Failed Tutorials of The Issue #6453 (#6534)

4 years agoAdd the SYSTEM keyword to all cmake include_directories commands for 3rd party or...
Robert Kimball [Wed, 23 Sep 2020 02:44:26 +0000 (19:44 -0700)]
Add the SYSTEM keyword to all cmake include_directories commands for 3rd party or external headers. Warning flags should only be applied to code within the tvm repository. (#6531)

4 years ago[Doc][Fix] Fix a typo in hybrid script tutorial. (#6525)
Zihao Ye [Wed, 23 Sep 2020 02:40:07 +0000 (10:40 +0800)]
[Doc][Fix] Fix a typo in hybrid script tutorial. (#6525)

4 years agoRemove settings about SGX in config.cmake (#6530)
qixiuai [Tue, 22 Sep 2020 17:07:48 +0000 (01:07 +0800)]
Remove settings about SGX in config.cmake (#6530)

removed settings about SGX since SGX is removed from TVM core

4 years ago[DOCS] Change some tutorial text (#6514)
Robert Kimball [Tue, 22 Sep 2020 14:47:34 +0000 (07:47 -0700)]
[DOCS] Change some tutorial text (#6514)

4 years agofix libtvm build dependencies when USE_MICRO is ON. (#6524)
Andrew Reusch [Tue, 22 Sep 2020 14:36:05 +0000 (07:36 -0700)]
fix libtvm build dependencies when USE_MICRO is ON. (#6524)

* previously, building from scratch would fail with Unix Makefiles
   due to cmake limitation

4 years ago[Torch] Clean up usage of try ... infer_value() ... except (#6504)
masahi [Tue, 22 Sep 2020 02:59:00 +0000 (11:59 +0900)]
[Torch] Clean up usage of try ... infer_value() ... except (#6504)

* clean up infer value usage

* try silence pylint

* remove unused variable

* make on_failuare optional

* make on_success optional True

Co-authored-by: masa <masa@pop-os.localdomain>
4 years ago[TOPI] Group conv2d NHWC op implementation (#6510)
Thierry Moreau [Tue, 22 Sep 2020 01:32:25 +0000 (18:32 -0700)]
[TOPI] Group conv2d NHWC op implementation (#6510)

4 years agoEnable more warnings when compiling with clang 10.0 or greater (#6456)
Robert Kimball [Tue, 22 Sep 2020 01:18:02 +0000 (18:18 -0700)]
Enable more warnings when compiling with clang 10.0 or greater (#6456)

4 years ago[AutoScheduler] Improve hyperlinks in the tutorial (#6521)
Lianmin Zheng [Tue, 22 Sep 2020 00:46:52 +0000 (17:46 -0700)]
[AutoScheduler] Improve hyperlinks in the tutorial (#6521)

* improve auto-scheduler tutorials

* improve tutorials

* fix lint

4 years ago[LINT] Use fmt off to disable problematic black fmt (#6519)
Tianqi Chen [Mon, 21 Sep 2020 15:46:41 +0000 (08:46 -0700)]
[LINT] Use fmt off to disable problematic black fmt (#6519)

4 years agoQnnBinaryLayout bugfix + unit test (#6513)
Dmitriy Smirnov [Mon, 21 Sep 2020 09:22:59 +0000 (10:22 +0100)]
QnnBinaryLayout bugfix + unit test (#6513)

An attempt to fix an issue which appeared when ConverLayout pass
runs on quantized binary operations like qnn.add:
def before():
    x = relay.var("x", shape=(2, 2))
    y = relay.var("y", shape=(1, 2))
    return relay.Function(
        [x, y],
        relay.qnn.op.add(
            x,
            y,
            lhs_scale=relay.const(0.0156863, "float32"),
            lhs_zero_point=relay.const(127, "int32"),
            rhs_scale=relay.const(0.0117647, "float32"),
            rhs_zero_point=relay.const(85, "int32"),
            output_scale=relay.const(0.0235294, "float32"),
            output_zero_point=relay.const(128, "int32"),
        ),
    )

The issue manifested itself as
  [bt] (2) ./src/tvm/build/libtvm.so(tvm::relay::qnn::QnnBinaryBroadcastLayout(tvm::Attrs const&, tvm::runtime::Array<tvm::tir::Layout, void> const&, tvm::runtime::Array<tvm::tir::Layout, void> const&, tvm::runtime::Array<tvm::Type, void> const&)+0xa9) [0x7fadc080d949]
  [bt] (1) ./src/tvm/build/libtvm.so(tvm::runtime::Array<tvm::tir::Layout, void>::operator[](long) const+0xb6) [0x7fadc0382996]
  [bt] (0) ./src/tvm/build/libtvm.so(+0xb50c12) [0x7fadc037cc12]
  File "/workspace/include/tvm/runtime/container.h", line 681
IndexError: Check failed: 0 <= i && i < p->size_: indexing 1 on an array of size 1

4 years ago[CI] Cancel previous build if new commit has been pushed to a PR (#6518)
Tianqi Chen [Mon, 21 Sep 2020 00:49:23 +0000 (17:49 -0700)]
[CI] Cancel previous build if new commit has been pushed to a PR (#6518)

4 years ago[TOPI] Fix declaration_conv2d_transpose_impl (#6428)
Xiao Xia [Sun, 20 Sep 2020 19:53:26 +0000 (03:53 +0800)]
[TOPI] Fix declaration_conv2d_transpose_impl (#6428)

4 years agoUpdate CI badge location (#6517)
Tianqi Chen [Sat, 19 Sep 2020 22:41:38 +0000 (15:41 -0700)]
Update CI badge location (#6517)

4 years ago[ANSOR] Auto-scheduler tutorial for GPU and necessary refactor/fix (#6512)
Lianmin Zheng [Sat, 19 Sep 2020 22:38:29 +0000 (15:38 -0700)]
[ANSOR] Auto-scheduler tutorial for GPU and necessary refactor/fix (#6512)

* add gpu tutorial

* refactor mutation in evolutionary search

* update

* update double matmul

* fix lint

* add double matmul test

* fix mutate compute location

* fix sketch search policy

* fix lint

* update

* address comments

* fix PruneInvalidStates

4 years ago[tvmc] command line driver 'compile' (part 2/4) (#6302)
Leandro Nunes [Fri, 18 Sep 2020 18:37:13 +0000 (19:37 +0100)]
[tvmc] command line driver 'compile' (part 2/4) (#6302)

* [tvmc] command line driver 'compile' (part 2/4)

 * Add 'compile' subcommand into tvmc (tvm.driver.tvmc)
 * Add frontends: Keras, ONNX, TensorFlow, tflite, PyTorch
 * Add tests for the 'compile' subcommand
 * Enable command line driver tests as part of integration tests
 * Skip tests if the cross-compilation toolchain is not installed

Co-authored-by: Marcus Shawcroft <marcus.shawcroft@arm.com>
Co-authored-by: Matthew Barrett <Matthew.Barrett@arm.com>
Co-authored-by: Dmitriy Smirnov <dmitriy.smirnov@arm.com>
Co-authored-by: Luke Hutton <luke.hutton@arm.com>
Co-authored-by: Giuseppe Rossini <giuseppe.rossini@arm.com>
Co-authored-by: Matthew Barrett <matthew.barrett@arm.com>
Co-authored-by: Elen Kalda <elen.kalda@arm.com>
Co-authored-by: Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
Co-authored-by: Jeremy Johnson <jeremy.johnson@arm.com>
Co-authored-by: Ina Dobreva <Ina.Dobreva@arm.com>
* tvmc: adjust TODOs

* tvmc: fix linting errors

* Address code-review comments

* Adjust pytest fixture to not break when there is no tensorflow

* Fix frontend tests, to cope with different frameworks in different images

* Apply suggestions from code review

Co-authored-by: Cody Yu <comaniac0422@gmail.com>
* Fix lint and code-review issues

* Re-format with black.

* tvmc: Move dependencies to extras_requires

Co-authored-by: Marcus Shawcroft <marcus.shawcroft@arm.com>
Co-authored-by: Matthew Barrett <Matthew.Barrett@arm.com>
Co-authored-by: Dmitriy Smirnov <dmitriy.smirnov@arm.com>
Co-authored-by: Luke Hutton <luke.hutton@arm.com>
Co-authored-by: Giuseppe Rossini <giuseppe.rossini@arm.com>
Co-authored-by: Elen Kalda <elen.kalda@arm.com>
Co-authored-by: Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
Co-authored-by: Jeremy Johnson <jeremy.johnson@arm.com>
Co-authored-by: Ina Dobreva <Ina.Dobreva@arm.com>
Co-authored-by: Cody Yu <comaniac0422@gmail.com>
4 years ago[RELAY][OP] roi_align operator alter layout (#6443)
Honghua Cao [Fri, 18 Sep 2020 17:29:12 +0000 (01:29 +0800)]
[RELAY][OP] roi_align operator alter layout (#6443)

* [RELAY][OP] roi_align operator alter layout

* [RELAY][OP] roi_align operator alter layout

* [RELAY][OP] roi_align operator alter layout

* [RELAY][OP] roi_align operator alter layout

Co-authored-by: honghua.cao <honghua.cao@streamcomputing.com>
4 years agoAdd beagleboard ai, thunderx and stm32mp1 to the arm_cpu target. (#6501)
Tom Gall [Fri, 18 Sep 2020 15:41:55 +0000 (10:41 -0500)]
Add beagleboard ai, thunderx and stm32mp1 to the arm_cpu target. (#6501)

* Add beagleboard ai, thunderx and stm32mp1 to the arm_cpu target.

Signed-off-by: Tom Gall <tom.gall@linaro.org>
* updates from black on target.py

Signed-off-by: Tom Gall <tom.gall@linaro.org>
4 years ago[Relay]Allow dynamic batch for arm conv2d (#6509)
Yao Wang [Fri, 18 Sep 2020 15:39:47 +0000 (08:39 -0700)]
[Relay]Allow dynamic batch for arm conv2d (#6509)

* Allow dynamic batch for arm conv2d

* Add TODO

4 years agoAdd several op mapping in PyTorch frontend (#6472)
Yong Wu [Fri, 18 Sep 2020 10:16:27 +0000 (18:16 +0800)]
Add several op mapping in PyTorch frontend (#6472)

* Add copy_ and clamp_ in PyTorch frontend

* add true_divide in PyTorch frontend

* more test cases for copy_

* fix format

* remove copy_

* fix format

* skip true_divide for torch < 1.5

4 years ago[COMMUNITY] hypercubestart -> Reviewer (#6511)
Yizhi Liu [Fri, 18 Sep 2020 08:34:40 +0000 (01:34 -0700)]
[COMMUNITY] hypercubestart -> Reviewer (#6511)

4 years agoimprove doc for relay.nn.dense (#6508)
Shizhi Tang [Fri, 18 Sep 2020 05:30:22 +0000 (13:30 +0800)]
improve doc for relay.nn.dense (#6508)

4 years ago[RPC] Lazily import micro when starting an RPC server (#6505)
lhutton1 [Thu, 17 Sep 2020 21:49:32 +0000 (22:49 +0100)]
[RPC] Lazily import micro when starting an RPC server (#6505)

* [RPC] Lazily import micro when starting an RPC server

Since #6334 the RPC server cannot be started unless USE_MICRO is enabled. I've tracked this down to an import in `python/tvn/exec/rpc_server.py`: `from tvm import micro` in the top level list of imports. This will mean that we try to import micro when it's not been built. Fix this by lazily importing micro when initializing an rpc server with micro enabled.

Change-Id: I8f22d81e215cfe4ac0662b0a99bdf02a3e91f90c

* fix lint

Change-Id: I8b78b678374bc82b3b66a7b3595ed4f1684e7d90

4 years ago[TIR][Hybrid] Hybrid Script Improvement (#6507)
Bohan Hou [Thu, 17 Sep 2020 21:49:09 +0000 (05:49 +0800)]
[TIR][Hybrid] Hybrid Script Improvement (#6507)

* [TIR][Hybrid] update

* [TIR][Hybrid] python formatting

4 years agoSwitch CRC-CCITT libraries (#6499)
Andrew Reusch [Thu, 17 Sep 2020 14:59:39 +0000 (07:59 -0700)]
Switch CRC-CCITT libraries (#6499)

4 years agoAdd PT OD tutorial (#6500)
Yao Wang [Thu, 17 Sep 2020 10:08:52 +0000 (03:08 -0700)]
Add PT OD tutorial (#6500)

4 years agofix a typo in topi key (#6502)
Shizhi Tang [Thu, 17 Sep 2020 10:08:13 +0000 (18:08 +0800)]
fix a typo in topi key (#6502)

4 years ago[Frontend][Pytorch] Improve Pytorch frontend for object detection models (#6449)
Yao Wang [Thu, 17 Sep 2020 03:32:20 +0000 (20:32 -0700)]
[Frontend][Pytorch] Improve Pytorch frontend for object detection models (#6449)

* Improve Pytorch Frontend

* Add tests

* Fix pylint

* Improve data cast

* Use int64 for slice axis

* Fix lint

* fix roi_align(..., aligned=True)

* Minor fix

* Add e2e test

* Add asf header

* Minor change

* Use dynamic topk

* Improve test

* Rollback topk

* py format

* remove print

* More improve

* Fix test

* Improve addmm

* Fix test

* Fix format

* Fix format

* Fix test scatter

Co-authored-by: q.yao <streetyao@live.com>
4 years ago[TUTORIAL][ANSOR] Using the template-free auto-scheduler on CPU (#6488)
Lianmin Zheng [Thu, 17 Sep 2020 02:14:09 +0000 (19:14 -0700)]
[TUTORIAL][ANSOR] Using the template-free auto-scheduler on CPU (#6488)

* add tutorial

* add tutorial

* update

* Apply suggestions from code review

Co-authored-by: Cody Yu <comaniac0422@gmail.com>
* address comments

* fix bugs

* add the exmple for resuming the search

* fix lint

Co-authored-by: Cody Yu <comaniac0422@gmail.com>
4 years agoextending FindVulkan to build RPC server on Windows correctly (#6498)
Thierry Moreau [Thu, 17 Sep 2020 01:11:55 +0000 (18:11 -0700)]
extending FindVulkan to build RPC server on Windows correctly (#6498)

Co-authored-by: Mei Ye <meiandmimi@yahoo.com>
Co-authored-by: Mei Ye <meiandmimi@yahoo.com>
4 years ago[BUILD] enhance build script for optional vta dep (#6497)
Tianqi Chen [Wed, 16 Sep 2020 23:19:52 +0000 (16:19 -0700)]
[BUILD] enhance build script for optional vta dep (#6497)

4 years agoblack format master (#6494)
Matthew Brookhart [Wed, 16 Sep 2020 22:37:36 +0000 (16:37 -0600)]
black format master (#6494)

4 years agobumping vta version (#6495)
Thierry Moreau [Wed, 16 Sep 2020 20:25:08 +0000 (13:25 -0700)]
bumping vta version (#6495)

4 years agoFix missing import in bifrost schedule (#6479)
Shizhi Tang [Wed, 16 Sep 2020 15:18:49 +0000 (23:18 +0800)]
Fix missing import in bifrost schedule (#6479)

4 years ago[PY] GraphRuntime: Update the tutorials to the module-based interface (#6482)
Tianqi Chen [Wed, 16 Sep 2020 15:18:06 +0000 (08:18 -0700)]
[PY] GraphRuntime: Update the tutorials to the module-based interface (#6482)

* [PY] GraphRuntime: Update the tutorials to the module-based interface.

Also added document about the encouraged usage.
In particular, we encourage the following usage.

lib = relay.build(...)
gmod = graph_runtime.GraphModule(lib["default"](ctx))

I have changed most of the tutorials and apps.
Some follow up PRs are needed to update some of the tests code.

* Fix VTA tutorials

4 years ago[Formatting] Fix python formatting issue (#6491)
lhutton1 [Wed, 16 Sep 2020 13:50:30 +0000 (14:50 +0100)]
[Formatting] Fix python formatting issue (#6491)

After #6442 was merged, black started complaining about formatting of test_pass_convert_op_layout.py. Format here.

Change-Id: I04346fa06e22b722b619488b895b47c6943e3fd9

4 years ago[BUG][ConvertLayout] Fix qnn.conv2d layout conversion too many values to unpack ...
lhutton1 [Wed, 16 Sep 2020 08:55:13 +0000 (09:55 +0100)]
[BUG][ConvertLayout] Fix qnn.conv2d layout conversion too many values to unpack (#6442)

This patch follows a previous bugfix in #6419. I made a very simple oversight for qnn.conv2d in that tinfos also contains qnn parameters. Therefore, we need to extract data_info and weight_info differently.

Change-Id: Ib0ad01f427543371380d0bb604a77b5e0ec1103d

4 years ago[FIX] Save tensor size with alignment (#6487)
Zhi [Wed, 16 Sep 2020 02:10:27 +0000 (19:10 -0700)]
[FIX] Save tensor size with alignment (#6487)

4 years ago[Minor] Fix typos in Ansor (#6425)
Junru Shao [Tue, 15 Sep 2020 23:36:10 +0000 (16:36 -0700)]
[Minor] Fix typos in Ansor (#6425)

4 years ago[BYOC][ETHOSN] Introduce further operator support (#6355)
mbaret [Tue, 15 Sep 2020 22:05:19 +0000 (23:05 +0100)]
[BYOC][ETHOSN] Introduce further operator support (#6355)

* [BYOC][ETHOSN] Introduce further operator support

This PR introduces support for the following operators:
 - Quantized Fully Connected
 - Quantized Addition
 - Depth-to-space
 - Max/Avg Pool 2D
 - Quantized Relu (Clip)
 - Reshape
 - Quantized Sigmoid

Co-authored-by: Leo Blonk <Leo.Blonk@arm.com>
Co-authored-by: Tristan O'Connor <tristan.oconnor@arm.com>
Co-authored-by: Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
* Skip tf imports if not available

Change-Id: I11bcf4a78014fa63e7b8e3b0cb00eecfd6cb7760

* ethos -> ethosn

Change-Id: I1fb1a11d0765f6d69f04c24b9c24e08665b8af6a

* Reduce random testing in test_addition

Change-Id: Id06063a0a0cf5f01356df23dc5d4bbbcb47cfa99

* Reduce random testing in test fullyconnected

Change-Id: I330408dfabc4bd804373f100581ce909ff724052

* Fix dumb mistake with rename

Change-Id: I2c5007be485b323116a0e8bab0f9106ea5ec834b

* Added comments to update the hashes in network tests when necessary

Change-Id: I13828c918c959daa492b9ed942a882c86d6690d1

* Fix github name

Change-Id: Idaa70ab9c2ec8db2828d51d15e7c23f28670ec82

* Use black formatting

Change-Id: I538171bd547a16395bef155a1dad28e8b3e347f2

Co-authored-by: Leo Blonk <Leo.Blonk@arm.com>
Co-authored-by: Tristan O'Connor <tristan.oconnor@arm.com>
Co-authored-by: Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
4 years agoadd git diff filter (#6484)
Cody Yu [Tue, 15 Sep 2020 20:19:59 +0000 (13:19 -0700)]
add git diff filter (#6484)

4 years agoµTVM RPC server and Part 1 of AutoTVM compilation infrastructure (#6334)
Andrew Reusch [Tue, 15 Sep 2020 16:10:26 +0000 (09:10 -0700)]
µTVM RPC server and Part 1 of AutoTVM compilation infrastructure (#6334)

4 years ago[DOC] Fix Some Broken Web Links (#6475)
Qiang Zhang [Tue, 15 Sep 2020 14:33:29 +0000 (22:33 +0800)]
[DOC] Fix Some Broken Web Links (#6475)

4 years ago[Relay]Some backend improvements for PT OD models (#6464)
Yao Wang [Tue, 15 Sep 2020 06:47:50 +0000 (23:47 -0700)]
[Relay]Some backend improvements for PT OD models (#6464)

* Some backend improve for PT od models

* Fix clang

* Fix pylint

* Enable gpu tests

* Minor fix

4 years ago[Formatting] Fix black script for Python formatting (#6469)
Cody Yu [Tue, 15 Sep 2020 00:17:18 +0000 (17:17 -0700)]
[Formatting] Fix black script for Python formatting (#6469)

4 years agoadd aten::pixel_shuffle implementation (#6328) (#6468)
wjliu [Mon, 14 Sep 2020 14:51:06 +0000 (22:51 +0800)]
add aten::pixel_shuffle implementation (#6328) (#6468)

4 years ago[RFC][Formatting] Add scripts for applying Black to the Python code. (#6437)
Jared Roesch [Sun, 13 Sep 2020 23:04:38 +0000 (16:04 -0700)]
[RFC][Formatting] Add scripts for applying Black to the Python code. (#6437)

4 years ago[ONNX] Update Slice op conversion to take strides into account, clean up tests (...
masahi [Sun, 13 Sep 2020 17:30:46 +0000 (02:30 +0900)]
[ONNX] Update Slice op conversion to take strides into account, clean up tests (#6467)

Co-authored-by: masa <masa@pop-os.localdomain>
4 years ago[CMAKE] Improve FindLLVM to handle llvm-prefix with space. (#6466)
Tianqi Chen [Sun, 13 Sep 2020 16:58:49 +0000 (09:58 -0700)]
[CMAKE] Improve FindLLVM to handle llvm-prefix with space. (#6466)

* [CMAKE] Improve FindLLVM to handle llvm-prefix with space.

Useful for windows settings where llvm can sits in Program Files.
Also updated the windows compiler to use clang.

* Additional updates

4 years ago[runtime] fix: remove anoymous namespace and rename BooleanToTranspose (#6465)
cloud-mxd [Sat, 12 Sep 2020 18:27:36 +0000 (02:27 +0800)]
[runtime] fix: remove anoymous namespace and rename BooleanToTranspose (#6465)

4 years ago[RPC] Update RPC module to enable remote linking. (#6462)
Tianqi Chen [Sat, 12 Sep 2020 18:27:18 +0000 (11:27 -0700)]
[RPC] Update RPC module to enable remote linking. (#6462)

Remote linking is useful when the linker is not available
on the host environment.

4 years ago[COMMUNITY] lhutton1 -> Reviewer (#6461)
Zhi [Sat, 12 Sep 2020 18:27:09 +0000 (11:27 -0700)]
[COMMUNITY] lhutton1 -> Reviewer (#6461)

4 years ago[COMMUNITY] comaniac -> Committer (#6463)
ziheng [Sat, 12 Sep 2020 18:26:26 +0000 (11:26 -0700)]
[COMMUNITY] comaniac -> Committer (#6463)

4 years ago[ONNX] Add support for GatherElements conversion (#6446)
masahi [Sat, 12 Sep 2020 12:44:15 +0000 (21:44 +0900)]
[ONNX] Add support for GatherElements conversion (#6446)

* support onnx GatherElements

* remove print

* run black

Co-authored-by: masa <masa@pop-os.localdomain>
4 years ago[APPS] Add How to deploy graph runtime example under new module factory (#6459)
Tianqi Chen [Sat, 12 Sep 2020 05:31:21 +0000 (22:31 -0700)]
[APPS] Add How to deploy graph runtime example under new module factory (#6459)

4 years ago[FIX,TESTING] Add tvm.testing to the docs (#6458)
Tristan Konolige [Sat, 12 Sep 2020 01:27:49 +0000 (18:27 -0700)]
[FIX,TESTING] Add tvm.testing to the docs (#6458)

4 years agoupstream (#6436)
雾雨魔理沙 [Sat, 12 Sep 2020 00:00:50 +0000 (17:00 -0700)]
upstream (#6436)

4 years ago[WINDOWS][MSVC] Fix MSVC warnings (#6450)
Tianqi Chen [Fri, 11 Sep 2020 20:15:13 +0000 (13:15 -0700)]
[WINDOWS][MSVC] Fix MSVC warnings (#6450)

* [WINDOWS][MSVC] Fix MSVC warnings

This PR fixes various warnings bought by MSVC.

TODO: deprecate `__tvm_main__` symbol and update
testcase so windows works as normal.

* Fix unicode problem in data_layout

4 years ago[topi, x86] for 1d loop, make outer loop parallel after split (#6455)
masahi [Fri, 11 Sep 2020 18:32:05 +0000 (03:32 +0900)]
[topi, x86] for 1d loop, make outer loop parallel after split (#6455)

Co-authored-by: masa <masa@pop-os.localdomain>
4 years ago[TEST] Fix Some Failed Test Cases and Tutorials of The Issue #6453 (#6454)
Johnson9009 [Fri, 11 Sep 2020 16:55:36 +0000 (00:55 +0800)]
[TEST] Fix Some Failed Test Cases and Tutorials of The Issue #6453 (#6454)

4 years ago[runtime] fix: BooleanToTranspose function definition conflict (#6452)
cloud-mxd [Fri, 11 Sep 2020 15:53:45 +0000 (23:53 +0800)]
[runtime] fix: BooleanToTranspose function definition conflict (#6452)

4 years agoAdd black to lint docker image (#6451)
Jared Roesch [Fri, 11 Sep 2020 14:45:41 +0000 (07:45 -0700)]
Add black to lint docker image (#6451)

4 years ago[Format] Convert all Python code w/o CI (#6448)
Jared Roesch [Fri, 11 Sep 2020 13:17:24 +0000 (06:17 -0700)]
[Format] Convert all Python code w/o CI (#6448)

* Add black setup

* Tweak pyproject.toml

* Fix syntax issues

* Fix

* Tweak

* Black all Python code

4 years agoROCm: use GcnArch for mcpu and ApiVersion to select code object version (#6447)
Thomas Viehmann [Fri, 11 Sep 2020 06:11:44 +0000 (08:11 +0200)]
ROCm: use GcnArch for mcpu and ApiVersion to select code object version (#6447)