Skip to content

[tvm4j] register user-defined function #251

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 9 commits into from
Jul 20, 2017
Merged

[tvm4j] register user-defined function #251

merged 9 commits into from
Jul 20, 2017

Conversation

@yzhliu
Copy link
Member

@yzhliu yzhliu commented Jul 16, 2017 •
edited
Loading

It is not ready to merge, while Function.main() works.

TBD:

  • callback function finalizer
  • unit test cases
  • support more argument types: bytes
  • support more argument types: NDArray

tqchen
tqchen reviewed Jul 16, 2017
View reviewed changes
jvm/native/src/main/native/ml_dmlc_tvm_native_c_api.cc Outdated
Copy link
Member

@tqchen tqchen Jul 16, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If it is important to move ownership to the java side before you return, see https://github.com/dmlc/tvm/blob/master/include/tvm/runtime/c_runtime_api.h#L269

Copy link
Member

@tqchen tqchen Jul 16, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

before you pass the arguments to the callback function, since callback funtion might want to retain the value

tqchen
tqchen reviewed Jul 16, 2017
View reviewed changes
jvm/native/src/main/native/jni_helper_func.h Outdated
Copy link
Member

@tqchen tqchen Jul 16, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Very likely you can reuse the same logic of implementing call in the front-end, since that is simply call into Return

Copy link
Member

@tqchen tqchen Jul 16, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

see https://github.com/dmlc/tvm/blob/master/web/tvm_runtime.js#L482

tqchen
tqchen reviewed Jul 16, 2017
View reviewed changes
jvm/native/src/main/native/jni_helper_func.h Outdated
Copy link
Member

@tqchen tqchen Jul 16, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There is one possible bug I think I find today, since resource won't be constructed until user call asModule.

If user call asModule twice, there will be double free. If user won;t call asModule in return value. There will be memory leak

@yzhliu
Copy link
Member Author

yzhliu commented Jul 20, 2017

@tqchen It is ready for review.

tqchen
tqchen reviewed Jul 20, 2017
View reviewed changes
jvm/core/src/main/java/ml/dmlc/tvm/Function.java Outdated
* @param override Whether override existing entry.
*/
public static void register(String name, Callback function, boolean override) {
int fid = genFunctionId(function);
Copy link
Member

@tqchen tqchen Jul 20, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Directly use jobject as handle as in https://github.com/dmlc/xgboost/blob/master/jvm-packages/xgboost4j/src/native/xgboost4j.cpp

Copy link
Member

@tqchen tqchen Jul 20, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

https://github.com/dmlc/xgboost/blob/master/jvm-packages/xgboost4j/src/native/xgboost4j.cpp#L44

tqchen
tqchen reviewed Jul 20, 2017
View reviewed changes
jvm/core/src/test/java/ml/dmlc/tvm/FunctionTest.java Outdated
}

@Test
public void test_reg_sum_ndarray() {
Copy link
Member

@tqchen tqchen Jul 20, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

To increase coverage, I would recommend to add the test that takes function as argument and returns a function

Copy link
Member

@tqchen tqchen Jul 20, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

c.f. https://github.com/tqchen/tvm/blob/master/tests/web/test_packed_func.js#L9

@tqchen tqchen merged commit 68c4400 into apache:master Jul 20, 2017
vinx13 pushed a commit to vinx13/tvm that referenced this pull request Mar 9, 2022
@MasterJH5574
[Typo] BufferRV (apache#251)
812ddce
apivovarov pushed a commit to apivovarov/tvm that referenced this pull request May 5, 2022
Release 1.10.1 stft (apache#251)
e3de8d7
mikepapadim pushed a commit to mikepapadim/tvm that referenced this pull request Oct 5, 2022
@MasterJH5574
[TVMScript] B2: match shape support (apache#251)
09527a8 
This PR features `match_shape` support, which contains:
- MatchShape support for both var emitting and non-emitting cases
- Allow TIR operators dtype broadcasting. (e.g. `i64 * i32`)

Co-authored-by: Ruihang Lai <[email protected]>
junrushao added a commit to junrushao/tvm that referenced this pull request Oct 18, 2022
@junrushao @YuchenJin @MasterJH5574
Recover dropped commits
985a123 
[TVMScript] B4: If branch support (apache#263)
B8: Local Function Support  (apache#258)
[TVMScript] B3: Type annotation checks (apache#256)
[TVMScript][Parser] B1: Dataflow block (apache#252)
[TVMScript] B2: match shape support (apache#251)
[TVMScript] B6/B7: Symbolic shape and var shadowing  (apache#245)
[TVMScript] B5: Support relax op (apache#244)
[TVMScript] B0: Call_tir support (apache#243)
enhance parser error reporting (apache#242)
[TVMScript] A1: Relax Parser infra (apache#240)
update ci image versions. (apache#241)
[TVMScript] B2-4: TIR IRBuilder (apache#239)
[TVMScript] A0: Relax IRBuilder infra (apache#235)
[TVMScript] B5-6: TIR IRBuilder (apache#231)
[TVMScript] B1: IRBuilder (apache#228)
[TVMScript] New Parser: Part C (apache#218)
[TVMScript] New Parser: Part A (apache#221)
[TVMScript] New Parser: Part B (apache#217)

Not recovered:
[Pass] Separate ApplyHistoryBest from tuning passes (apache#226)
[Bugfix] Couple of bug fixes to run TVM-gen code together with BYOC (apache#249)

co-authored-by: Yuchen Jin <[email protected]>
co-authored-by: Siyuan Feng <[email protected]>
co-authored-by: Ruihang Lai <[email protected]>
MasterJH5574 added a commit to MasterJH5574/tvm that referenced this pull request Nov 20, 2022
@junrushao @YuchenJin @MasterJH5574
Recover dropped commits
f9f4f79 
[TVMScript] B4: If branch support (apache#263)
B8: Local Function Support  (apache#258)
[TVMScript] B3: Type annotation checks (apache#256)
[TVMScript][Parser] B1: Dataflow block (apache#252)
[TVMScript] B2: match shape support (apache#251)
[TVMScript] B6/B7: Symbolic shape and var shadowing  (apache#245)
[TVMScript] B5: Support relax op (apache#244)
[TVMScript] B0: Call_tir support (apache#243)
enhance parser error reporting (apache#242)
[TVMScript] A1: Relax Parser infra (apache#240)
update ci image versions. (apache#241)
[TVMScript] B2-4: TIR IRBuilder (apache#239)
[TVMScript] A0: Relax IRBuilder infra (apache#235)
[TVMScript] B5-6: TIR IRBuilder (apache#231)
[TVMScript] B1: IRBuilder (apache#228)
[TVMScript] New Parser: Part C (apache#218)
[TVMScript] New Parser: Part A (apache#221)
[TVMScript] New Parser: Part B (apache#217)

Not recovered:
[Pass] Separate ApplyHistoryBest from tuning passes (apache#226)
[Bugfix] Couple of bug fixes to run TVM-gen code together with BYOC (apache#249)

co-authored-by: Yuchen Jin <[email protected]>
co-authored-by: Siyuan Feng <[email protected]>
co-authored-by: Ruihang Lai <[email protected]>
junrushao added a commit to junrushao/tvm that referenced this pull request Feb 8, 2023
@junrushao @YuchenJin @MasterJH5574
Recover dropped commits
8b90ee0 
[TVMScript] B4: If branch support (apache#263)
B8: Local Function Support  (apache#258)
[TVMScript] B3: Type annotation checks (apache#256)
[TVMScript][Parser] B1: Dataflow block (apache#252)
[TVMScript] B2: match shape support (apache#251)
[TVMScript] B6/B7: Symbolic shape and var shadowing  (apache#245)
[TVMScript] B5: Support relax op (apache#244)
[TVMScript] B0: Call_tir support (apache#243)
enhance parser error reporting (apache#242)
[TVMScript] A1: Relax Parser infra (apache#240)
update ci image versions. (apache#241)
[TVMScript] B2-4: TIR IRBuilder (apache#239)
[TVMScript] A0: Relax IRBuilder infra (apache#235)
[TVMScript] B5-6: TIR IRBuilder (apache#231)
[TVMScript] B1: IRBuilder (apache#228)
[TVMScript] New Parser: Part C (apache#218)
[TVMScript] New Parser: Part A (apache#221)
[TVMScript] New Parser: Part B (apache#217)

Not recovered:
[Pass] Separate ApplyHistoryBest from tuning passes (apache#226)
[Bugfix] Couple of bug fixes to run TVM-gen code together with BYOC (apache#249)

co-authored-by: Yuchen Jin <[email protected]>
co-authored-by: Siyuan Feng <[email protected]>
co-authored-by: Ruihang Lai <[email protected]>
yelite pushed a commit to yelite/tvm that referenced this pull request Feb 17, 2023
@junrushao @YuchenJin @MasterJH5574
Recover dropped commits
9e6acd0 
[TVMScript] B4: If branch support (apache#263)
B8: Local Function Support  (apache#258)
[TVMScript] B3: Type annotation checks (apache#256)
[TVMScript][Parser] B1: Dataflow block (apache#252)
[TVMScript] B2: match shape support (apache#251)
[TVMScript] B6/B7: Symbolic shape and var shadowing  (apache#245)
[TVMScript] B5: Support relax op (apache#244)
[TVMScript] B0: Call_tir support (apache#243)
enhance parser error reporting (apache#242)
[TVMScript] A1: Relax Parser infra (apache#240)
update ci image versions. (apache#241)
[TVMScript] B2-4: TIR IRBuilder (apache#239)
[TVMScript] A0: Relax IRBuilder infra (apache#235)
[TVMScript] B5-6: TIR IRBuilder (apache#231)
[TVMScript] B1: IRBuilder (apache#228)
[TVMScript] New Parser: Part C (apache#218)
[TVMScript] New Parser: Part A (apache#221)
[TVMScript] New Parser: Part B (apache#217)

Not recovered:
[Pass] Separate ApplyHistoryBest from tuning passes (apache#226)
[Bugfix] Couple of bug fixes to run TVM-gen code together with BYOC (apache#249)

co-authored-by: Yuchen Jin <[email protected]>
co-authored-by: Siyuan Feng <[email protected]>
co-authored-by: Ruihang Lai <[email protected]>
MasterJH5574 added a commit to MasterJH5574/tvm that referenced this pull request Jun 14, 2023
@MasterJH5574
[Analysis] Reshape TIR detection with iter-map-simplify (apache#251)
e7d10e5 
This PR supports the reshape PrimFunc detection analysis with
iter-map-simplify, in order to better support TIR functions with dynamic
variables, which is not well handled before.

Meanwhie, this PR also keeps the existing analysis logic, so that we are
still able to cover the cases like reshape with raggedness.

Unit tests for reshape are added to ensure correctness.
MasterJH5574 pushed a commit to MasterJH5574/tvm that referenced this pull request Aug 17, 2025
@Kathryn-cat @jinhongyii
kernel: gemm-rs overlap, using dynamic scheduler (apache#251)
91b00a9 
* 1.79us

Co-authored-by: Hongyi Jin <[email protected]>

* 1.78us

---------

Co-authored-by: Hongyi Jin <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@tqchen tqchen tqchen approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@yzhliu @tqchen