Hua [Tue, 4 Jun 2019 16:47:29 +0000 (09:47 -0700)]
[Bugfix] [VTA] VTA DRAM Have A Logic Issue May Cause GEMM Output Wrong. (#3278)
* [Bugfix] [VTA] VTA DRAM Have A Logic Issue May Cause GEMM Output Wrong.
Symptom:
after change “LOG_BLOCK_IN” and “LOG_BLOCK_OUT” from vta_config.json
into 7, run vta "Simple Matrix Multiply" in "simulator", the vta
calculate result for GEMM is wrong.
Sometime VTA crash with error “Check failed: phy_addr != 0 (0 vs. 0) :
trying to get address that is nullptr”
Analysis:
Simulator hardcode kPageSize into 1<<12 and physical address calculate
based on this size, when doing “insn->dram_base” calculation , because
GetElemBytes(dst_memory_type) larger than page size, different physcial
address may get same dram_base, than caused logic issue and finally
trigger GEMM out put is wrong.
Solution:
add logic to check if PAGE SIZE larger then "GetElemBytes" return value.
* address review comments.
Jared Roesch [Tue, 4 Jun 2019 15:42:47 +0000 (08:42 -0700)]
Bump ONNX version (#3286)
Sergei Grechanik [Tue, 4 Jun 2019 15:42:27 +0000 (18:42 +0300)]
[ARITH] Bugfix: int bound analysis for mod (#3288)
Zhi [Mon, 3 Jun 2019 17:40:38 +0000 (10:40 -0700)]
[RELAY][TRANSFORM] Migrate buildmodule to transform (#3251)
Sergei Grechanik [Mon, 3 Jun 2019 15:52:31 +0000 (18:52 +0300)]
[ARITH] Bugfix: check arg positiveness for mod rules (#3279)
Alexander Pivovarov [Sat, 1 Jun 2019 18:16:16 +0000 (11:16 -0700)]
Update tflite tutorial to use TFLite r1.13 schema (#3271)
Zhi [Sat, 1 Jun 2019 07:53:18 +0000 (00:53 -0700)]
[relay][heterogeneous] annotate using visitor (#3261)
* annotate using visitor
* retrigger CI
Hua [Sat, 1 Jun 2019 02:42:15 +0000 (19:42 -0700)]
[Bugfix][VTA] PkgConfig cause crash in PYNQ board due to link library (#3257)
* [Bugfix][VTA] PkgConfig cause crash in PYNQ board due to link library
not exist.
Symptom:
When run vta_get_started.py with pynq board, host crash and
complain "cannot find -lsds_lib" and "cannot find -l:libdma.so"
Reproduce:
At pynq board, delete the ./build/vta_config.json, then run rpc
server.
In host machine run vta_get_started.py, issue would reproduce.
Analysis:
This issue caused by 'PkgConfig' function still using pynq2.1
library which not exist in pynq2.4 anymore, when a "reconfig_runtime"
logic of rpc_server.py get triggered , the compile would failed due to
link library not exist.
Solution:
change the link library to libcma.so.
* [Document Change][VTA] Change pynq version from 2.3 into 2.4.
Issue:
pynq 2.3 image not available anymore from pynq download page and pynq
2.4 is the current latest image which available in the said website, after
verification, currently VTA work good with pynq 2.4 image, hence update
related document from pynq 2.3 to 2.4.
Logan Weber [Fri, 31 May 2019 19:21:36 +0000 (12:21 -0700)]
Enable uTVM in Jenkinsfile (#3269)
Animesh Jain [Fri, 31 May 2019 08:29:54 +0000 (01:29 -0700)]
[Relay][Hashing] Structural hash - incorporate the var type into its hash (#3267)
Currently, the BindVar function does not take Var type into account. This causes
two same graph structures with different var shapes to have same hash.
Structural hash is used for keeping track of which operators we have
already compiled. Because of this, two operators with different shapes end up
pointing to same compiled code. The failure is encountered at runtime, where the
expected input shape asserts are not met.
Tianqi Chen [Fri, 31 May 2019 04:32:33 +0000 (21:32 -0700)]
Jekyll (#3262)
hlu1 [Fri, 31 May 2019 04:11:25 +0000 (21:11 -0700)]
[Bugfix] Fix a memory leak in OpManager (#3263)
Balint Cristian [Fri, 31 May 2019 02:12:56 +0000 (05:12 +0300)]
[Relay] Handle float16 constants & fix BatchNorm (#3260)
Yao Wang [Wed, 29 May 2019 23:36:05 +0000 (16:36 -0700)]
[AutoTVM]Core functionality for Graph tuner (#2184)
* Add graph tuning
* Add tests
* Fix tests
* Fix pylint
* Small fix for docstring
* Minor fix
* Support fetching workload from relay expr
* Simplify benchmark layout transformation
* Add relay support
* Fix infer layout func name
* Refactor internal data representation
* Fix issues
* Add PBQP solver
* Fix layout transform check
* Add PBQPTuner test
* Fix lint
* Update tutorial
* Fix tutorial
* Fix lint
* Add relay test
* Remove nnvm since nnvm graph can be converted to relay function
* Modify benchmark layout wrt new layout_transform api
* Fix lint
* Update docstring for DP tuner
* Refactor traverse graph
* Support graph tuning for multiple target operators
* Fix fetching workloads
* Add x86 depthwise_conv2d infer_layout
* Fix x86 depthwise_conv2d autotvm
* Fix PBQP tuner
* Fix DP tuner
* Generate dummy layout transform record
* Update tutorial
* Modify layout records name
* Add ASF header
* Add ASF header for testing files
* Fix test
* Fix topi fetching
* Some refactors
* Fix lint
* Fix tutorial
* Rename test files
* Fix doc typo
* Add test case note link
Hua [Wed, 29 May 2019 17:32:47 +0000 (10:32 -0700)]
[BugFix][VTA] Fix vta_conv2d crash issue after change vta_config.json configuration. (#3213)
Issue:
Once change LOG_BLOCK_IN or LOG_BLOCK_OUT into > 4 value, when run vta
“Simple Matrix Multiply” or load vta, vta would crash at vta_conv2d.py.
Analysis:
This issue caused by resnet18 logic of vta_conv2d.py which have
in_filter minmum size that is 16. > 4 value would cause such in_filter
check failed then make xfer_size be empty and find_schedules function
return a empty list finally cause crash.
Solution:
add the empty list check.
Tianqi Chen [Wed, 29 May 2019 01:12:17 +0000 (18:12 -0700)]
[C++] Cleanup transform API nits (#3253)
masahi [Tue, 28 May 2019 22:20:58 +0000 (07:20 +0900)]
[TOPI] Fix resize nearest with fractional scaling (#3244)
Nick Hynes [Tue, 28 May 2019 22:20:18 +0000 (15:20 -0700)]
[RUST] Rust DSO module (#2976)
Gus Smith [Tue, 28 May 2019 20:35:25 +0000 (13:35 -0700)]
Typo: Tensorflow --> TensorFlow (#3249)
Gus Smith [Tue, 28 May 2019 20:35:09 +0000 (13:35 -0700)]
kCustomBegin overlapped with kExtEnd; incr by 1 (#3250)
This was a typo in the original custom datatypes PR.
Wuwei Lin [Tue, 28 May 2019 18:02:48 +0000 (02:02 +0800)]
Move CombineParallelConv2D to opt level 4 (#3248)
Luis Vega [Tue, 28 May 2019 04:29:55 +0000 (21:29 -0700)]
[VTA][TSIM] Use Module instead of RawModule for testbench by creating an empty bundle for the IO (#3242)
* use Module instead of RawModule for testbench by creating an empty bundle for the IO
* change default back to verilog
Wei Chen [Mon, 27 May 2019 22:24:54 +0000 (06:24 +0800)]
[Doc][Relay] Add VM doc (#3188)
* [Doc][Relay] Add VM doc
* Add Apache header
* Apply suggestions from code review
Co-Authored-By: Steven S. Lyubomirsky <slyubomirsky@gmail.com>
Co-Authored-By: 雾雨魔理沙 <lolisa@marisa.moe>
Co-Authored-By: Logan Weber <36520469+weberlo@users.noreply.github.com>
Co-Authored-By: Zhi <5145158+zhiics@users.noreply.github.com>
* Junru's comment
* More fix
* More fix
* More fix
* last fix
* Apply suggestions from code review
Co-Authored-By: 雾雨魔理沙 <lolisa@marisa.moe>
* Apply suggestions from code review
Co-Authored-By: Logan Weber <36520469+weberlo@users.noreply.github.com>
* Add code links
* Remove unused bp
* Update docs/dev/virtual_machine.rst
Co-Authored-By: Logan Weber <36520469+weberlo@users.noreply.github.com>
* Explain TODO
* Yong's comment
Co-Authored-By: Yong Wu <55wuyong@163.com>
* Comment
Sergei Grechanik [Mon, 27 May 2019 16:33:13 +0000 (19:33 +0300)]
[ARITH] Improve div/mod in rewrite simplifier (#3149)
* [ARITH] Improve div/mod in rewrite simplifier
* Fix lint error
* Fuller file name in src/arithmetic/modular_set.h
Co-Authored-By: Wei Chen <ipondering.weic@gmail.com>
* Generalize some rules
* Replace gcd factoring with specialized rules
* Mark rules that don't work for non-truncated division
* More tests
Haichen Shen [Sun, 26 May 2019 00:40:02 +0000 (17:40 -0700)]
[Relay][Frontend] Add Crop op converter (#3241)
* Add Crop op converter
* lint
* x
Gemfield [Sat, 25 May 2019 00:48:58 +0000 (08:48 +0800)]
Fixed a typo (#3218)
* Fixed a typo
* Remove outdated url link.
雾雨魔理沙 [Fri, 24 May 2019 23:43:03 +0000 (16:43 -0700)]
[Relay] Start porting pass to the pass manager (#3191)
雾雨魔理沙 [Fri, 24 May 2019 23:42:29 +0000 (16:42 -0700)]
[Relay] remove unneeded VisitExpr (#3239)
Siju [Fri, 24 May 2019 21:38:08 +0000 (03:08 +0530)]
[RELAY]Frontend darknet (#2773)
* [RELAY]Frontend darknet
* CI test file updated & CI error fixed
* avg_pool pad fix
* Changed repo_url and doc formatting
Zhi [Fri, 24 May 2019 19:05:00 +0000 (12:05 -0700)]
[Relay][Transform] merge PassContext and BuildConfig (#3234)
Tianqi Chen [Fri, 24 May 2019 16:29:14 +0000 (09:29 -0700)]
[C++][API] Consistent RAII scoping API. (#3231)
eqy [Fri, 24 May 2019 00:52:44 +0000 (17:52 -0700)]
[LINT] handle more file types in ASF header (#3235)
* Update add_asf_header.py
* Update add_asf_header.py
Josh Fromm [Thu, 23 May 2019 17:23:12 +0000 (10:23 -0700)]
Modified pick best to accumulate the best configurations from both the input and output file. (#3225)
Tianqi Chen [Thu, 23 May 2019 17:17:03 +0000 (10:17 -0700)]
[NODE] Macro to define NodeRef methods, constructor style example (#3224)
hlu1 [Thu, 23 May 2019 17:13:11 +0000 (10:13 -0700)]
[GraphRuntime] Debug graph runtime (#3232)
Bing Xu [Wed, 22 May 2019 23:22:13 +0000 (16:22 -0700)]
[3rdparty] sync submodules (#3229)
Gus Smith [Wed, 22 May 2019 22:41:02 +0000 (15:41 -0700)]
Register SkipVectorize (#3228)
Steven S. Lyubomirsky [Wed, 22 May 2019 20:57:53 +0000 (13:57 -0700)]
[Relay][Prelude] Remove Peano nats from the prelude (#3045)
Zhi [Wed, 22 May 2019 20:52:52 +0000 (13:52 -0700)]
[relay][pass manager] Open transform namespace (#3226)
Yuta Hinokuma [Wed, 22 May 2019 18:48:39 +0000 (03:48 +0900)]
[WIP] [Relay] [NNVM] [Frontend] implement MaxPool-8 and MaxPool-10 (#3114)
Leyuan Wang [Wed, 22 May 2019 18:09:01 +0000 (11:09 -0700)]
[Bugfix] Fix sort changing original input data issue (#3212)
* sort bugfix for not rearranging input data
* separate sort schedule
* fix lint
* use identity op instead
* fix lint
* remove redundent code
llyfacebook [Wed, 22 May 2019 16:11:17 +0000 (09:11 -0700)]
Add packing for int8 1x1 convolution and support the int8 group convolution on X86 (#2991)
* Support the 1x1 int8 conv with NHWC layout and weight packing
fix linter
* fix the memoize issue
* fix the failed nhwc test
* add the schedule for pack to unbreak other tests
* skip avx512 compile
* Support the 1x1 int8 conv with NHWC layout and weight packing
fix linter
* fix the memoize issue
* fix the failed nhwc test
* add the schedule for pack to unbreak other tests
* skip avx512 compile
* Unify the data_layout and kernel_layout relation
* add asf header
* fix the comment
* retrigger the build/test
Yizhi Liu [Wed, 22 May 2019 05:28:49 +0000 (22:28 -0700)]
[Team] Eddie -> PMC (#3220)
Zhao Wu [Wed, 22 May 2019 02:44:47 +0000 (10:44 +0800)]
[TFLite] Convert TFLite NCHW to NHWC (#3141)
* Convert TFLite NCHW to NHWC
* Minor comment fix
Logan Weber [Tue, 21 May 2019 23:34:35 +0000 (16:34 -0700)]
Add `SkipVectorize` pass (#3222)
hlu1 [Tue, 21 May 2019 23:05:28 +0000 (16:05 -0700)]
[Contrib] cblas batch_matmul (#3210)
Zhi [Tue, 21 May 2019 19:53:58 +0000 (12:53 -0700)]
[Relay][heterogeneous pass] remove on_device op after annotation (#3204)
* remove on_device op after annotation
* Update src/relay/pass/device_annotation.cc
Co-Authored-By: MORINAGA <34588258+imorinaga@users.noreply.github.com>
Zhi [Tue, 21 May 2019 19:53:03 +0000 (12:53 -0700)]
[relay][vm] remove throw in destructor (#3215)
Haichen Shen [Tue, 21 May 2019 08:03:34 +0000 (01:03 -0700)]
Add bing to reviewer (#3214)
Yong Wu [Mon, 20 May 2019 18:56:22 +0000 (11:56 -0700)]
[Relay][TOPI] operator All (#3124)
* [Relay][TOPI] operator All
* Update tests/python/frontend/tensorflow/test_forward.py
Co-Authored-By: yongwww <55wuyong@163.com>
* fix comments
* change to level 4
Joshua Z. Zhang [Mon, 20 May 2019 17:07:32 +0000 (10:07 -0700)]
[Relay][ONNX] fix #3134 converter where initializers were not registered as nodes (#3143)
Haichen Shen [Mon, 20 May 2019 17:07:01 +0000 (10:07 -0700)]
[BugFix] Fix bug in cast to bool (#3207)
Luis Vega [Mon, 20 May 2019 17:05:36 +0000 (10:05 -0700)]
[VTA] [TSIM] Improve tsim example (#3206)
Hua [Mon, 20 May 2019 17:01:49 +0000 (10:01 -0700)]
[BugFix][VTA] Fix bug in vta runtime DepPop function. (#3208)
Issue:
One of existing illegal dependency check's condition always true,
the correct logic actually should be such check for store and load.
Solution:
Fix the said logic issue.
hlu1 [Fri, 17 May 2019 17:29:07 +0000 (10:29 -0700)]
[ARM] Fix concat (#3061)
lixiaoquan [Fri, 17 May 2019 17:13:17 +0000 (01:13 +0800)]
[CODEGEN][CUDA][OPENCL] Handle INF and NAN (#3194)
Josh Fromm [Fri, 17 May 2019 10:41:50 +0000 (03:41 -0700)]
[Relay] Better shape inference in TensorFlow Frontend. (#3176)
* Some bug fixes in tensorflow graph converter and added DepthToSpace operator.
* Made DepthToSpace better comply with other function syntax.
* Added better shape inference for unusual situations.
* Lint fixes.
* Added depthtospace test.
* Added test cases for value inference and depthtospace.
* Added fill testing.
* Made comment changes and added BroadcastTo op and tests.
* Fixed underlining and unneeded opt_level forcing.
* Added _infer_value assertion that all values to infer are available in passed parameters.
Tianqi Chen [Thu, 16 May 2019 20:22:31 +0000 (13:22 -0700)]
[RELAY] Hotfix build_module creation (#3198)
雾雨魔理沙 [Thu, 16 May 2019 18:54:43 +0000 (11:54 -0700)]
Update .gitignore (#3199)
Philipp Krones [Thu, 16 May 2019 18:00:38 +0000 (20:00 +0200)]
Quick fix of VTA FPGA Toolchain Installation documentation (#3196)
Wuwei Lin [Thu, 16 May 2019 15:39:13 +0000 (23:39 +0800)]
[TOPI] Raise exception group_conv2d_nchw not supported (#3195)
Siva [Thu, 16 May 2019 03:55:38 +0000 (09:25 +0530)]
[TENSORLFOW] PlaceholderWithDefault (limited) implementation. (#3184)
Mark Rogers [Thu, 16 May 2019 03:26:24 +0000 (20:26 -0700)]
Get list of unsupported ONNX operators (#2995)
llyfacebook [Thu, 16 May 2019 03:21:35 +0000 (20:21 -0700)]
Add the acc16 intrinsic support (#3081)
Josh Fromm [Thu, 16 May 2019 02:03:40 +0000 (19:03 -0700)]
[Relay] Option to select which convolution layers are quantized. (#3173)
* Stashing for later maybe.
* Added new option to leave specific layers unquantized.
* Better error checking.
* remove unneeded import
* tab to spaces
* pylint fixes
* more pylint fixes
Zhi [Thu, 16 May 2019 00:28:18 +0000 (17:28 -0700)]
[Relay][Compilation] replace relay.build_module with C++ BuildModule (#3174)
Gus Smith [Wed, 15 May 2019 20:34:30 +0000 (13:34 -0700)]
[Datatypes] Custom datatypes (#2900)
* Register and use custom datatypes in TVM
This patch adds the ability to register and use a custom datatype from Python,
using the `register_datatype` call. The datatype can then be passed as the
`dtype` parameter using the syntax `dtype="custom[<type_name>]bitsxlanes"`.
* Removes extra file
* Register custom datatypes with TVM; specify Cast and Add lowering
This commit adds functionality for registering custom datatypes with TVM, and
furthermore adding custom lowering functions to lower those custom datatypes.
This commit only adds lowering for the Cast and Add ops; more ops will be added
soon.
Check out some custom datatype samples in my repository of samples:
https://github.com/gussmith23/tvm-custom-datatype-samples
* Register and lower casts from Python
* Formatting
* Fix include; was including too much
* Add comment
* Add DatatypeRegistered
* Add storage size field to custom datatypes
This field indicates the bitwidth of the opaque block of data into which
instances of the datatype will be stored, when TVM compiles. For example, if I
create a datatype with a storage size of 16, then
- Constants of that datatype will be created as unsigned 16-bit ints
- Calls to external functions taking that datatype will pass the data as
unsigned 16-bit ints
- External functions returning that datatype will be assumed to return unsigned
16-bit ints.
* Change how lowering funcs (Cast and other ops) are named in registry
tvm.datatypes.lower.<target>.cast.<dst-type>.<src-type>
becomes
tvm.datatypes.lower.<target>.Cast.<dst-type>.<src-type>
And fixes some sloppy code around how the other ops were being formatted.
* Update Python register_datatype to accept storage size
* Oops, left out one cast->Cast change
* Look up storage size when parsing `custom[typename]`
When we encounter this type string in Python, it will be parsed into a Halide
type object in C++. Some of my original code supported this parsing, but we now
have to attach the storage type to the type (by setting the bits field).
* Change how external calls for casting/other ops are done
Firstly, we now use the storage size of the custom type when determining
input/output types; e.g. a cast to a custom type with storage size 16 is seen as
a call to an external function returning an opaque uint of size 16.
Secondly, write a macro to handle the other ops. Originally I thought I could
handle these at runtime, with a single `_register_op` global. I transitioned
instead to using individual `_register_Add` etc. calls generated with a macro,
but I don't remember why.
* When encountering a custom type immediate, generate UIntImm
* Translate custom types to LLVM type
* Generate correct return type in Casts
Originally I was assuming that the result type from casts was always a custom
datatype, and so I was making the Call return a UInt type.
* Use TVM-idiomatic recursion style in DatatypesLowerer
This was actually a bug, I'm pretty sure; we wouldn't have recursed deep on any
complex programs. As a result of making this change, I also uncovered another
potential bug, where the datatypes lowering pass would attempt to lower a Load
of a custom type. By commenting out the `Mutate_` for Load, I was able to stop
the error from cropping up, but frankly, I'm not satisfied with the solution;
how is it that we are able to run codegen when Loads of custom datatypes are
present in the IR? I have not written any code, to my knowledge, that will
support this. Perhaps Load does not care about the underlying datatype?
* Use CHECK
* Add comment about which Mutate_s are needed
* Add comments
* Add GetCustomDatatypeRegistered as an extern C function
* Formatting, comments, casting
* Change how datatype string is formatted
* Use bits() instead of GetStorageSize
Use bits() instead of GetStorageSize
* Change comment
* Add datatype.py
* Change registered function name (datatypes->datatype)
* Remove GetStorageSize
* Format custom datatypes like any other datatype
Specifically, we now print the bits and lanes after the `custom[...]` string.
* Correctly implement datatype lowering in Python
* Remove unneeded include
* Make function naming consistent
* Use CHECK instead of internal_assert
* Rename macro
* Formatting
* Rename functions
* Implement Cast lowering
`_datatype_register_op` is now able to lower both binary ops and Casts.
* Formatting
* Formatting
* Clang format, google style
* Fix std::string/extern "C" warnings
* Formatting
* Formatting
* Lower Allocates and Loads during datatype lowering
This should ensure that there are no custom datatypes remaining once datatype
lowering is done. This will allow us to remove the code in the LLVM codegen
which deals with custom datatypes.
* Revert additions to codegen_llvm.cc which are now unneeded
* Pass cpplint on lower_datatypes.cc
* Add clarifying comment
* Remove datatype lowering registration funcs from C++
* Add CHECKs
* Remove TODO
* Remove all references to storage size
* Move and rename function
* Rename function
* Remove done TODOs and other handled comments
* Remove irrelevant Load code and comments
* Comment out the IR node types I'm not sure about yet
* Add bfloat16 datatype unittest
* Fix MakeConstScalar
MakeConstScalar for a custom datatype will now call out to a function which can
be registered on a per-datatype basis. The function will take a double and
return the equivalent value in the custom datatype format.
Note that these code paths are not actually used or tested at the moment. I have
not yet written an example which uses const scalars of a custom datatype.
* Formatting
* Change pass name
* Allow users to register whatever lowering function they want
Tianqi pointed out that users should be able to register whatever lowering
function they want, and should not be constrained to registering lowering
functions which just call out to external libraries.
I still provide a function for making lowering functions which call out to
external libraries, for convenience.
* Add clarifying comment
* Remove unneeded comment
* Remove unneeded function
* Rename file
* Undo unnecessary change
* Undo unnecessary change
* Make naming consistent
Rename "datatypes" to "custom datatypes" in most contexts.
* Revert an artifact of old code
* Fix build warnings, add TODO
* Lint
* Remove unnecessary use of extern C by separating decl and impl
* Error checking
* Remove TODO
* Missed a name change
* Lint
* Python lint
* Correctly format datatype
* Move bfloat16 to 3rdparty
* "custom_datatypes" --> "datatype" in most places
I left the pass as "LowerCustomDatatypes" to indicate that we're not lowering
anything other than custom datatypes. Otherwise, everything else has been
changed.
* Upgrade datatype unittest
I used a float calculator to generate some real testcases for the unittest.
* Separate public includes and private implementation
Specifically, create cleaner decoupling between datatypes stuff in packed_func
and the datatype registry implementation.
* Formatting
* Limit custom datatype codes to >128
* Add TODOs
* Fix comment
* Formatting
* Clean up datatype unittest
* Remove un-exported functions in public headers; UIntImm->FloatImm
More places where I accidentally was using implementation-only functions in
public headers.
Additionally, store custom datatype immediates as FloatImms. A later change will
add new lowering logic to lower these FloatImms to UIntImms.
Plus formatting change.
* Lint
* Use FloatImm (not UIntImm) to hold immediates of custom datatypes
This change switches from using UIntImm to FloatImm for storing immediates of
custom datatypes. The value of the number is stored in a double, which should be
enough precision for now, for most custom types we will explore in the immediate
future.
In line with this change, we change the datatype lowering so that FloatImms are
lowered to UInts of the appropriate size. Originally, this was going to be done
by allowing the user to register a double->uint_<storage size>_t conversion
which would be called at compile time to convert the value from the FloatImm to
a UInt and store it in a UIntImm. After discussions with Tianqi, we decided to
take the simpler route, and lower FloatImms just as we lower all other ops: by
replacing them with Call nodes. In this case, presumably the user will Call out
to a conversion function in their datatype library.
The justification for this decision is due to the functionality added in #1486.
This pull request adds the ability to load LLVM bytecode in at compile time.
This applies in our case as follows:
1. The user writes their custom datatype programs and registers their lowering
functions in the same way we've been doing it so far. All operations over
custom datatypes are lowered to Calls to the datatype library.
2. The user compiles their datatype library to LLVM bytecode.
3. At TVM compile time, the user loads the LLVM bytecode. Depending on how the
datatype library is written, Clang should be able to perform constant
folding over the custom datatype immediates, even if their conversions are
done with calls to the library.
Additionally adds test to test the FloatImm codepath.
* Re-add a change I removed accidentally during rebase
* Cleanup
* Remove unnecessary TVM_DLLs
* Add custom datatype utilities source file to Go runtime pack
* Revert "Remove unnecessary TVM_DLLs"
This reverts commit
4b742b99557fd3bf0ce6617f033c8b444b74eda4.
* Mark bfloat code as TVM_DLL
* Moves custom datatype runtime utilities to c_runtime_api.cc
* Revert "Add custom datatype utilities source file to Go runtime pack"
This reverts commit
aecbcde0b2cc09a2693955b77037fe20f93b5bfd.
* Move datatype parsing to its own function
* Change comments
* Remove unneeded function
* Formatting
* Formatting
* Documentation
* Add kCustomBegin, use it for checking for custom types
* Documentation
* Formatting
* Move static definition to implementation
* Remove comment
* Decide toBeLowered before lowering arguments of Expr
In the past, e.g. when lowering custom datatypes for an Add, we would lower a
and b first, and then decide whether the resulting new Add needed to be lowered
based on the (new) types of a and b. Now, instead, we need to check the types of
a and b first (to see if they're custom types), and then lower them (so they'll
become non-custom types), and then lower the new Add.
* Revert "Move datatype parsing to its own function"
This reverts commit
d554a5881afcf69af1c070d882a7651022703a09.
This broke parsing. Will figure this out later. There isn't a really clean way
to separate this out given how the rest of the function is written.
* Replace comment
* Documentation
* Remove comment and TVM_DLL
* Better error messages
* Remove artifact of rebase
* Separate datatypes parsing to its own function
* Add \returns
* Comment changes; add TODO
* Refactor tests
Yong Wu [Wed, 15 May 2019 05:42:34 +0000 (22:42 -0700)]
[Relay][TensorFlow Frontend] SoftPlus Sqrt (#3187)
ghostplant [Tue, 14 May 2019 17:22:33 +0000 (01:22 +0800)]
Avoid using heavy API to query single attribution (#3179)
eqy [Tue, 14 May 2019 12:34:16 +0000 (05:34 -0700)]
[RELAY][PASS] detect depthwise conv2d in mac_count pass (#3083)
* check in
* use groups
* CHECK_EQ
* trigger CI
* Update mac_count.cc
* trigger CI
* trigger CI
Joshua Z. Zhang [Mon, 13 May 2019 21:17:11 +0000 (14:17 -0700)]
add onnx elemwise greater/less (#3186)
Salem Derisavi [Mon, 13 May 2019 19:48:00 +0000 (15:48 -0400)]
cleanup: removed a piece of code that is redundant now given updates to HalideIR submodule (#3169)
Oldpan [Mon, 13 May 2019 18:03:41 +0000 (02:03 +0800)]
Fix a bug of flatten in ONNX to Relay converter (#3180)
* fix onnx frontend flatten bug
* Update onnx.py
* Update onnx.py
* Update onnx.py
Siva [Mon, 13 May 2019 17:03:32 +0000 (22:33 +0530)]
[GOLANG] Some fixes for golang latest version compiler. #3119 (#3182)
Jared Roesch [Sat, 11 May 2019 22:08:13 +0000 (18:08 -0400)]
[Relay][Runtime] Add VM compiler. (#3139)
* Implement the VM compiler
* Fix issues
* Fix ASF headers
* Fix test issue
* Apply typo fixes.
* Update src/relay/backend/vm/compiler.cc
Co-Authored-By: 雾雨魔理沙 <lolisa@marisa.moe>
* Refactor compiler
* Fix
* Fix
* Fix in benchmark
* Fix
* Address comments
hlu1 [Sat, 11 May 2019 05:55:11 +0000 (22:55 -0700)]
[Bugfix] Check file exists before removing it (#3178)
lixiaoquan [Sat, 11 May 2019 04:56:01 +0000 (12:56 +0800)]
[Relay][TensorFlow] Support tf.math.reduce_prod (#3166)
Steven S. Lyubomirsky [Sat, 11 May 2019 00:25:54 +0000 (17:25 -0700)]
Register all operators' Python attributes in Python so they can be easily accessed from Python code (#3175)
Lianmin Zheng [Fri, 10 May 2019 23:36:54 +0000 (07:36 +0800)]
[HybridScript] Capture constant external python variables (#3157)
lixiaoquan [Fri, 10 May 2019 17:14:39 +0000 (01:14 +0800)]
Fix a tensorflow test bug. (#3165)
Length of input_shape isn't always 4.
Zhi [Fri, 10 May 2019 04:29:16 +0000 (21:29 -0700)]
[codegen] heterogeneous build for c++ (#3144)
* heterogeneous build for c++
* merge relay buildmodule to codegen build
* use module split
* use target_host
* remove sse3
* retrigger ci
Yinghai Lu [Thu, 9 May 2019 15:46:10 +0000 (08:46 -0700)]
Fix a multithreaded bug in llvm LazyInitJIT (#3158)
Leyuan Wang [Thu, 9 May 2019 15:44:47 +0000 (08:44 -0700)]
add more syncs (#3151)
Jared Roesch [Thu, 9 May 2019 06:09:15 +0000 (02:09 -0400)]
[Relay][Runtime] Implementation of Relay VM (#2889)
* Implement the virtual machine
Co-Authored-By: wweic <ipondering.weic@gmail.com>
* Fix rebase build issues
* Reorganize vm.py and fix allocator bug
* Remove compiler
* Remove tests
* Remove backend/vm/vm.cc too
* Fix docs
* Fix doc
* Fix doc
* Add vm docs
* Remove change to dead_code.cc
* Remove Relay logging
* Remove reduce
* Update include/tvm/runtime/vm.h
Co-Authored-By: jroesch <roeschinc@gmail.com>
* Reformat
* Update include/tvm/runtime/vm.h
Co-Authored-By: jroesch <roeschinc@gmail.com>
* Address feedback
* Update include/tvm/runtime/vm.h
Co-Authored-By: jroesch <roeschinc@gmail.com>
* Apply suggestions from code review
Co-Authored-By: jroesch <roeschinc@gmail.com>
* Fix a couple outstanding comments
* Last couple comments
* Update include/tvm/runtime/vm.h
Co-Authored-By: jroesch <roeschinc@gmail.com>
* Address code review feedback
* Fix final comment
* Address comments
* Error reporting and example
* add Const
* Explicitly delete copy assignment operator
* Fix rebase
* Pass 3rd arg to fusion
Bing Xu [Thu, 9 May 2019 04:06:33 +0000 (21:06 -0700)]
[BuildModule] Fix AlterLayout Pass (#3155)
Yao Wang [Thu, 9 May 2019 00:21:41 +0000 (17:21 -0700)]
[Relay][Op] Adaptive pooling (#3085)
* Add topi adaptive_pool
* Use adaptive_pool to compute global_pool
* Add relay adaptive pool2d
* Fix lint
* Fix typo
* Minor change
* Change support level to 10
* Add contrib
* Remove global pool schedule
* Add contrib module
* Fix lint
* Update doc
* Update doc
Luis Vega [Wed, 8 May 2019 16:59:00 +0000 (09:59 -0700)]
[RFC] [VTA] [TSIM] Enabling Cycle-Accurate Hardware Simulation for VTA #3009 (#3010)
* merge files
* move verilator to the right place
* change name to tsim
* add default rule to be build and run
* add README for tsim
* Update README.md
* add some structural feedback
* change name of VTASim to VTADPISim
* more renaming
* update comment
* add license
* fix indentation
* add switch for vta-tsim
* add more licenses
* update readme
* address some of the new feedback
* add some feedback from cpplint
* add one more whitespace
* pass pointer so linter is happy
* pass pointer so linter is happy
* README moved to vta documentation
* create types for dpi functions, so they can be handle easily
* fix pointer style
* add feedback from docs
* parametrize width data and pointers
* fix comments
* fix comment
* add comment to class
* add missing parameters
* move README back to tsim example
* add feedback
* add more comments and remove un-necessary argument in finish
* update comments
* fix cpplint
* fix doc
Zhi [Wed, 8 May 2019 15:23:10 +0000 (08:23 -0700)]
fix python lint warnings (#3145)
Tianqi Chen [Wed, 8 May 2019 15:14:08 +0000 (08:14 -0700)]
[CI] Always run cpptest during build to ensure library correctness (#3147)
Bing Xu [Wed, 8 May 2019 07:16:15 +0000 (00:16 -0700)]
Relay C++ Build Module (#3082)
* [Relay] C++ Build module
* asdf
Leyuan Wang [Wed, 8 May 2019 04:51:00 +0000 (21:51 -0700)]
[Bugfix][TOPI] conv2d_transpose bugfix (#3138)
* deconv tests
* deconv bug fixed for certain cases tests added
Marcus Shawcroft [Wed, 8 May 2019 04:50:44 +0000 (05:50 +0100)]
[DOC] fix :code: markup syntax (#3140)
Wei Chen [Wed, 8 May 2019 03:52:24 +0000 (20:52 -0700)]
Handle vectorize for LE statement (#3137)
* Handle vectorize for LE statement
Fix a new cases introduced by commit
7afbca5691fdb599cd90b043d5a5036e55cae2d6
* Add test
masahi [Sun, 5 May 2019 12:17:29 +0000 (21:17 +0900)]
[ROCm] Fix dense autotvm template registration (#3136)
* Fix rocm dense autotvm template
* suppres lint warning
Yong Wu [Sun, 5 May 2019 08:08:10 +0000 (01:08 -0700)]
[Relay][Frontend] add log op in tf frontend (#3111)
* [Relay][Frontend] add log op in tf frontend
* address comment
Lianmin Zheng [Sun, 5 May 2019 08:05:18 +0000 (16:05 +0800)]
[TOPI] Fix mali conv2d performance regression (#3131)
* [TOPI] fix mali conv
* fix typo
* address comments
Tianqi Chen [Sat, 4 May 2019 01:07:14 +0000 (21:07 -0400)]
[ARITH] Constraint-aware ConstIntBound, Enhance CanonicalSimplify (#3132)
Jessica Davies [Fri, 3 May 2019 09:48:18 +0000 (11:48 +0200)]
[DOC] Developer documentation for InferBound pass. (#3126)
* Developer documentation for InferBound pass.
Marcus Shawcroft [Thu, 2 May 2019 16:11:37 +0000 (17:11 +0100)]
[DOC] Various documentation improvements (#3133)
Jared Roesch [Thu, 2 May 2019 16:10:34 +0000 (12:10 -0400)]
[Relay][Runtime] Add memory manager for NDArray (#3121)
* Add support for custom NDArray memory management
Credit to @icemelon9 and @wweic
* Fix copy-paste issue
* Fix naive allocator.h
* Remove buffer field
* Apply Wei's suggestions.
Co-Authored-By: jroesch <roeschinc@gmail.com>
* Fix Wei's suggestion
* Fix go rts
* Break MM dependency
* Add docs and clean up diff
* Add more docs
* Move to VM folder
* Fix lint
* Remove Go dep.
* Rename to Empty
* Address Haichen's comments