[UMA] UMA v1.0 #12087

MichaelJKlaiber · 2022-07-14T08:47:04Z

UMA v1:

UMA Partitioner
UMA Pipeline
UMA Lower
UMA Codegen

cbalint13

@MichaelJKlaiber , @cgerum , @PaulPalomeroBernardo

My congratulations for the wonderful work and the outstanding idea towards modularization !

Suggested a small nit for cmake completeness
Please rebase against apache:main HEAD, it looks out of sync
Please invoke with Cc more reviewers (i.e. those from the UMA RFC PR)

CMakeLists.txt

cbalint13 · 2022-07-15T23:02:29Z

Cc @driazati may help with the missing clang i386 test failure @ test_lower_with_uma[1-224-224-3-5-5-4]

[2022-07-15T13:33:16.200Z] >  raise RuntimeError("cannot find clang, candidates are: " + str(cc_list))

sunggg

Thank you for the great contribution for better accelerator supports! 🥳
I enjoyed reading the code and learning the design.
Overall, implementation looks good to me.
Here are a few high-level comments:

Could you add more documentation in general (e.g., explain what parameters are)?
It would be great to have a testing for end-to-end model-level testing. (e.g., run BERT with UMA on ARM EthosU)

Please see below for more comments. Thanks!

python/tvm/relay/backend/contrib/uma/_template/backend.py

sunggg · 2022-07-18T22:31:50Z

python/tvm/relay/backend/contrib/uma/api/codegen.py

+
+    def _register_c_codegen(
+        self,
+        includes: Callable[[], str] = None,


Could you add the docstring for parameters?
Also, can includes be Optional[str]?

Doc-string added.
@cgerum : can you answer the other question?

To generate dynamic strings more easily we would be in favor of passing a Callable
@PaulPalomeroBernardo @cgerum

src/relay/backend/contrib/uma/tir_to_runtime.cc

python/tvm/relay/backend/contrib/uma/api/partitioner.py

python/tvm/relay/backend/contrib/uma/api/utils.py

cbalint13 · 2022-07-19T10:50:31Z

Thank you for the great contribution for better accelerator supports! partying_face I enjoyed reading the code and learning the design. Overall, implementation looks good to me. Here are a few high-level comments:

Could you add more documentation in general (e.g., explain what parameters are)?

It would be great to have a testing for end-to-end model-level testing. (e.g., run BERT with UMA on ARM EthosU)

Please see below for more comments. Thanks!

Thank you for the great contribution for better accelerator supports! partying_face I enjoyed reading the code and learning the design. Overall, implementation looks good to me. Here are a few high-level comments:

Personal view over the first part (non code related) of @sunggg requests, of course just IMHO:

Could you add more documentation in general (e.g., explain what parameters are)?

Current sample (_template) & doc are fine for a contrib module.
It greatly enables "out-of-tree" maintenance towards custom HW accelarator design.
Example parts Vanilla, Strawberry & Chocolate are nice levels of introductory & guidance.

It would be great to have a testing for end-to-end model-level testing. (e.g., run BERT with UMA on ARM EthosU)

Sounds beyond of scope, same way we could "dream" a fully verilated BERT running verilator-contrib module (!) @tmoreau89 , @vegaluisjose
As for ARM-Ethos ⟨™⟩, my understanding is a quite different (not so "out-of-tree" principled) way to connect things inside, but @manupa-arm can comment more on this.
Would be possible in the future to decouple ARM-Ethos⟨™⟩ (or Vitis-AI) to be more "out-of-tree" by porting it over UMA, but this would be quite bit of another task.

Some personal efforts envisioned similar out-of-tree approaches {for i.e. OLIMP / e-verest } are before UMA just landed as super nice "out-of-tree" topping. As my last wish, I put forward hope that we community one day will be able to leverage jointly the complex hardware-design space within TVM's unique compiler-space.

Beyond code-related requests below, I halt my review here with a @lgtm claim, and a go for UMA v1.0 contrib !

Please see below for more comments. Thanks!

sunggg · 2022-07-19T15:42:43Z

@cbalint13, appreciate sharing your thoughts! I also think this is a great out-of-tree approach and look forward to its actual usage.

Could you add more documentation in general (e.g., explain what parameters are)?

Current sample (_template) & doc are fine for a contrib module.
It greatly enables "out-of-tree" maintenance towards custom HW accelarator design.
Example parts Vanilla, Strawberry & Chocolate are nice levels of introductory & guidance.

I agree that the overall docstring structure (e.g., Vanilla, Strawberry & Chocolate) would be a great introduction/tutorial (Truly appreciate the authors!).
My issue was more like a code-level docstring since some of them are missing and I wasn't sure what each function or parameter exactly means.
https://github.com/apache/tvm/blob/bc211fdc91932ef9b77f7edd2ff65151ee5faaff/python/tvm/relay/backend/contrib/uma/_template/codegen.py
https://github.com/apache/tvm/blob/bc211fdc91932ef9b77f7edd2ff65151ee5faaff/python/tvm/relay/backend/contrib/uma/_template/codegen.py

It would be great to have a testing for end-to-end model-level testing. (e.g., run BERT with UMA on ARM EthosU)

Sounds beyond of scope, same way we could "dream" a fully verilated BERT running verilator-contrib module (!) @tmoreau89 , @vegaluisjose

As for ARM-Ethos ⟨™⟩, my understanding is a quite different (not so "out-of-tree" principled) way to connect things inside, but @manupa-arm can comment more on this.

Would be possible in the future to decouple ARM-Ethos⟨™⟩ (or Vitis-AI) to be more "out-of-tree" by porting it over UMA, but this would be quite bit of another task.

I see. Thank you for the clarification! Maybe my example was not relevant. I just thought it would be nice to have some model-level test cases since we seem to have all necessary components (UMA Partitioner, UMA Pipeline, UMA Lower, UMA Codegen) in this PR. I just wanted to see if current PR also covers a case where we lower some parts with UMA while handling other unsupported parts via the conventional pipeline.

areusch

thanks @MichaelJKlaiber I've given this a look over. I think it's generally in the right direction, suggest a couple more comments and a few minor changes here and there. overall though, the approach looks great. i'd love it if we could explore integrating this further with tvmc since we are registering a custom Target. happy to discuss further!

tests/python/contrib/test_uma/test_uma_lowering_with_umalower.py

cmake/config.cmake

include/tvm/relay/transform.h

python/tvm/relay/backend/contrib/uma/_template/backend.py

tests/python/contrib/test_uma/test_partition.py

python/tvm/relay/backend/contrib/uma/tutorial.md

areusch · 2022-07-20T05:05:16Z

src/relay/backend/contrib/uma/relay_to_tir.cc

+ * "Compiler" attribute. Functions that do not match this condition remain
+ * unaltered.
+ */
+class OutlineCompilerFunctionsMutator : public MixedModeMutator {


i think this already exists in ethosu, can we factor it out instead of duplicating?

cc : @lhutton1 , If its exactly the same and if you feel it useful in a general sense for accelerator flow. I would like to second @areusch, we could move this pass to be generic transform pass.

this is in fact just a copy of the ethosu code. I think, it is a generally useful pass for accelerator flows, so +1 to make this a generic transform pass, if this is an option.

@areusch, should we move this to a seperate PR then?

since we're duplicating it here, i'd prefer if we could factor it out now

ok, i'm fine to defer this to a follow-on PR if @manupa-arm is ok with that.

Maybe with a TODO/issue then ?

src/relay/backend/contrib/uma/relay_to_tir.cc

src/relay/backend/contrib/uma/tir_to_runtime.cc

python/tvm/relay/backend/contrib/uma/tutorial.md

cbalint13 · 2022-07-20T09:10:51Z

thanks @MichaelJKlaiber I've given this a look over. I think it's generally in the right direction, suggest a couple more comments and a few minor changes here and there. overall though, the approach looks great. i'd love it if we could explore integrating this further with tvmc since we are registering a custom Target. happy to discuss further!

@areusch ,
IMHO, for consideration purpose:

UMA with it's approach to bridge "out-of-tree" things/implementations into TVM might have "looser" tie & offerings for tvmc. I would only second the idea that tvmc's "exposing fixed CLI features" must relay when in reality it only may reach what (full/partially) will ever be implemented (by UMA consumers) on top of UMA bridge itself. And this may statement is valid even if we register a full tvm legit custom target via UMA.
I can imagine someone's only goal via UMA is to generate lowered "c" code blobs and period, code that includes super-optimised (inc. auto-tensorized loops) __asm(".word " 0xc0fefe) calls for very special ISA extensions. I won't condition this PR to tvmc with a must however some really obvious things probably can be exposed via tvmc.

MichaelJKlaiber · 2022-07-20T13:05:11Z

@cbalint13, appreciate sharing your thoughts! I also think this is a great out-of-tree approach and look forward to its actual usage.

Could you add more documentation in general (e.g., explain what parameters are)?

Current sample (_template) & doc are fine for a contrib module.
It greatly enables "out-of-tree" maintenance towards custom HW accelarator design.
Example parts Vanilla, Strawberry & Chocolate are nice levels of introductory & guidance.

I agree that the overall docstring structure (e.g., Vanilla, Strawberry & Chocolate) would be a great introduction/tutorial (Truly appreciate the authors!). My issue was more like a code-level docstring since some of them are missing and I wasn't sure what each function or parameter exactly means. https://github.com/apache/tvm/blob/bc211fdc91932ef9b77f7edd2ff65151ee5faaff/python/tvm/relay/backend/contrib/uma/_template/codegen.py https://github.com/apache/tvm/blob/bc211fdc91932ef9b77f7edd2ff65151ee5faaff/python/tvm/relay/backend/contrib/uma/_template/codegen.py

It would be great to have a testing for end-to-end model-level testing. (e.g., run BERT with UMA on ARM EthosU)

Sounds beyond of scope, same way we could "dream" a fully verilated BERT running verilator-contrib module (!) @tmoreau89 , @vegaluisjose

As for ARM-Ethos ⟨™⟩, my understanding is a quite different (not so "out-of-tree" principled) way to connect things inside, but @manupa-arm can comment more on this.

Would be possible in the future to decouple ARM-Ethos⟨™⟩ (or Vitis-AI) to be more "out-of-tree" by porting it over UMA, but this would be quite bit of another task.

I see. Thank you for the clarification! Maybe my example was not relevant. I just thought it would be nice to have some model-level test cases since we seem to have all necessary components (UMA Partitioner, UMA Pipeline, UMA Lower, UMA Codegen) in this PR. I just wanted to see if current PR also covers a case where we lower some parts with UMA while handling other unsupported parts via the conventional pipeline.

We have an example testcase for running MobileNet (created in Relay) with uma and the Vanilla Accelerator in place. Not sure if it should be added to this PR or a later one, as this PR already is a lot to discuss. I'd be in favor of moving this to the next UMA PR.

manupak

Thanks for the great work!
I've done an intial round of review -- my concerns are mostly around the structure and UMA CLI. It would be better seperate the latter to another PR, IMO, given the content in this PR.

python/tvm/relay/backend/contrib/uma/uma_cli.py

manupak · 2022-07-20T14:23:41Z

python/tvm/relay/backend/contrib/uma/_template/run.py

+    AOTTestRunner,
+    generate_ref_data,
+    compile_and_run,
+)


This feels like a test source. Should we move this to tests ?

@manupa-arm , this is not supposed to be test code. This is example code for the tutorial. We will replace AOTTestRunner by microTVM Project API

python/tvm/relay/backend/contrib/uma/_template/backend.py

manupak · 2022-07-20T14:33:27Z

src/relay/backend/contrib/uma/relay_to_tir.cc

+ * "Compiler" attribute. Functions that do not match this condition remain
+ * unaltered.
+ */
+class OutlineCompilerFunctionsMutator : public MixedModeMutator {


cc : @lhutton1 , If its exactly the same and if you feel it useful in a general sense for accelerator flow. I would like to second @areusch, we could move this pass to be generic transform pass.

python/tvm/relay/backend/contrib/uma/_template/backend.py

lhutton1

Thanks @MichaelJKlaiber this is an exciting addition! Just wanted to chip in with a few more suggestions / nits.

Since the code in python/tvm/relay/backend/contrib/uma/* is a developer facing API, it might be worth enabling mypy type checking in task_mypy.sh.

python/tvm/relay/backend/contrib/uma/api/partitioner.py

lhutton1 · 2022-07-20T15:03:17Z

python/tvm/relay/backend/contrib/uma/api/lower.py

+                ],
+                Optional[int],
+            ]
+        ] = []


Would it make sense to break these type annotations into smaller named variables? e.g. like PatternTable in partitioner.py

just noting i think this one isn't done yet

python/tvm/relay/backend/contrib/uma/backend.py

tests/scripts/task_python_uma.sh

tests/python/contrib/test_uma/test_uma_vanilla_accelerator.py

tests/python/contrib/test_uma/test_partition.py

areusch · 2022-07-21T21:38:52Z

we discussed this in the Community Meeting yesterday. here are notes on the discussion:

when the design references "Accelerator A" and "Accelerator B," does this mean we're using both simultaneously?
- not in this v1, though the architecture supports it. at present they can simply coexist as options.
should we integrate this with TVMC?
- @areusch: it should be fairly easy to integrate the UMA targets with the tvmc run command
- @manupa-arm : this should be pretty straightforward to add to tvmc. the bigger concern here was around uma_cli.py, which is supposed to generate a starter implementation for new accelerators in uma.
- @areusch : we should either have tvmc or some other developer-facing entry point to house tools like this. probably not bad to add dev tools to tvmc now--we can always migrate them out if we need to.
- @MichaelJKlaiber : intention of uma_cli is just to make the tutorial easier to replicate on your own, so there are two steps there--create the accelerator flow and then run inference.
- @manupa-arm : do we expect the CLI to work when we're in an environment where only the tvm wheel is present? e.g what about the C sources included with accelerator? should those go in the wheel?
- @MichaelJKlaiber: those sources are copied into the generated dir by uma_cli.
- @areusch : what's the include path folks are expected to set on their downstream C compiler? seems like the C files included with accelerator template should really make it into the Model Library Format. Could produce another CSourceModule which would create another e.g. default_lib3.cc in the MLF. Could also use the import_c pragma similar. to how we do for microTVM.
where should the template live?
- @areusch : could go either way or both. how do we expect people to package their accelerator flow? if merging into mainline, perhaps we want in the python import path. if keeping accelerator flow private, perhaps apps is similar to carrying that code alongside the tvm wheel.
- @manupa-arm : deciding intended location based on whether a flow will get upstreamed makes sense. _template is an example rather than a target, so maybe apps could make more sense for it.
@manupa-arm : also suggest to break the CLI changes into another PR.
@MichaelJKlaiber : only vanilla accelerator was impl'd; do folks have suggestions for chocolate and strawberry? feel free ot post in discuss thread or contact
- @areusch : would be cool to see something that leverages usmp to model physical accelerator memories. could also be cool to see an example where buffers were marked to live on-device.
Slava: are the optimization provided in the default TVM pipeline also part of the UMA pipeline?
- @areusch : you can classify the optimizations in terms of relay passes, scheduling, and post-scheduling passes. TVM tries to operate on an IRModule-to-IRModule prinicple, where each optimization or step takes an IRModule and returns an IRModule. when you mark a subgraph as offloaded to an UMA pipeline, some optimizations aren't enabled--for example, Relay-level operator fusion. Others e.g. those which operate post-scheduling (usmp, for example) will run on UMA operators.
- Slava: if I have a conv2d followed by batch norm, and only the conv2d is offloaded, then the batch norm is not fused by default?
- @areusch: the right way to do that would be to mark both as offloaded and do the fusion yourself. there are also some efforts to enable post-scheduling fusion via Relax, but those haven't landed yet.
Slava: what's the best way to leverage UMA if e.g. we have 2 different implementations of conv2d depending on kernel size?
- @areusch : you'd need to give your pattern matcher enough fidelity to differentiate those two workloads. you can also inspect the matched subgraph after using a looser pattern.
slava: what's the rough timeline?
- not really a timeline, but see [Tracking Issue] UMA: Universal Modular Accelerator Interface #11260
@MichaelJKlaiber : can also discuss more questions in high-bandwidth with folks.
- suggest folks post up on the discuss forum. we can also use this meeting for further discussion.

UMAv1.0 updates according to comments and suggestions from PR apache#12087

areusch

thanks @MichaelJKlaiber! few last things

areusch · 2022-07-27T16:07:49Z

src/relay/backend/contrib/uma/relay_to_tir.cc

+ * "Compiler" attribute. Functions that do not match this condition remain
+ * unaltered.
+ */
+class OutlineCompilerFunctionsMutator : public MixedModeMutator {


since we're duplicating it here, i'd prefer if we could factor it out now

tests/python/contrib/test_uma/test_partition.py

src/relay/backend/contrib/uma/relay_to_tir.cc

areusch · 2022-07-27T17:30:04Z

python/tvm/relay/backend/contrib/uma/_template/backend.py

+        #######################################################################
+        # Relay to TIR function registration
+        #######################################################################
+        self._register_tir_pass(PassPhase.TIR_PHASE_0, MyAiHwConv2dPass())


cc @mbaret one thing we've been discussing is that it might make sense for these enums to roughly correspond to "guarantees about the form of the IRModule" (e.g. "all the primitive Functions are lowered" or "Buffers are defined"). To that end I think it would be great to avoid referencing e.g. TIR_PHASE_n in the enum unless those are distinct guarantees.

python/tvm/relay/backend/contrib/uma/_template/passes.py

MichaelJKlaiber · 2022-08-01T09:05:49Z

CI pipeline still fails for the target i386, see https://ci.tlcpack.ai/blue/rest/organizations/jenkins/pipelines/tvm/branches/PR-12087/runs/12/nodes/360/steps/1207/log/?start=0
As this does not seem an issue introduced by UMA code, I will remove UMA for i386 build, i.e.

tests/scripts/task_config_build_i386.sh
echo set(USE_UMA OFF) >> config.cmake

What are your thoughts?
@cbalint13 @areusch @areusch @cgerum @manupa-arm @PaulPalomeroBernardo

Does anyone of you understand what is different from i386 to build_cpu?

Does it have to do with:
echo set(USE_LLVM llvm-config-4.0) >> config.cmake
VS
echo set(USE_LLVM llvm-config-11) >> config.cmake
?

MichaelJKlaiber · 2022-08-03T07:42:38Z

Thanks to all the reviewers for the thorough reviews and the great suggestions and comment that improved UMA!
@manupa-arm @areusch @sunggg @cbalint13 @kslavka @lhutton1

We integrated the vast majority of the requests in this review into the PR.
There were many great suggestion that we hadn't considered before, but improved the overall quality of the UMA contribution, e.g.

Tutorial as RST for gallery
Adding a case test case where the TVM default pipeline and the UMA pipeline are both used
...

This PR has been open for a quite long time now, it has improved steadily and we feel that it is ready now. Please let us know you thoughts.

There are a couple of (small) things we left out for the is PR, and would ask for you opinion to move it to (very near future) PRs:

move OutlineCompilerFunctionsMutator out of UMA @areusch @manupa-arm
replace AOTTestRunner by project API @manupa-arm
annotations of UMALower._operator_strategies @lhutton1
(these steps will be moved to the tracking issue, for the case that the PR is granted)

Further steps see current tracking issue #11260

It is also great that WE as a community identified items where further discussion is required, e.g.:

UMA CLI / tvmc integration
Demo use-cases, e.g. running BERT on Ethos-U with UMA
...
I'd move them to an updated version of the UMA RFC or the discussion forum

Thanks to everyone contributing to UMA, esp @cgerum @PaulPalomeroBernardo

CC: @SebastianBoblest @UlrikHjort

areusch

thanks @MichaelJKlaiber ! I found a few small mostly mechanical fixes that would make this PR perfect, but overall it looks good to me. thanks for your (and everyone else involved) hard work on this!

apps/uma/_template/__init__.py

apps/uma/_template/conv2dnchw.cc

apps/uma/_template/passes.py

areusch · 2022-08-04T15:43:11Z

python/tvm/relay/backend/contrib/uma/api/lower.py

+                ],
+                Optional[int],
+            ]
+        ] = []


just noting i think this one isn't done yet

areusch · 2022-08-04T15:45:17Z

src/relay/backend/contrib/uma/relay_to_tir.cc

+ * "Compiler" attribute. Functions that do not match this condition remain
+ * unaltered.
+ */
+class OutlineCompilerFunctionsMutator : public MixedModeMutator {


ok, i'm fine to defer this to a follow-on PR if @manupa-arm is ok with that.

sunggg

Thank you for the exciting contribution and addressing my comments!
Just a super minor nit :)

tests/python/contrib/test_uma/test_uma_pipeline.py

manupak

Thanks for the all the hard work here and for all the changes.
I had a deeper look into the PR. Few nits and a concern with "includes" function for UMABackend.

manupak · 2022-08-05T08:07:54Z

apps/uma/_template/conv2dnchw.cc

+     *
+     */
+    int
+    my_ai_hw_conv2dnchw(float* ifmap, float* weights, float* result, int oc, int iw, int ih, int ic,


This source looks like as it was not linted. Maybe we dont lint apps ..
Anyhow, would you be able to run clang-format on this file ?

I agree that it doesnt look linted. It is what
./docker/lint.sh -i clang_format
gives me

@leandron @areusch @driazati do you know any reason that this file will not be linted ?

include/tvm/relay/transform.h

python/tvm/relay/backend/contrib/uma/api/lower.py

manupak · 2022-08-05T08:21:12Z

src/relay/backend/contrib/uma/relay_to_tir.cc

+ * "Compiler" attribute. Functions that do not match this condition remain
+ * unaltered.
+ */
+class OutlineCompilerFunctionsMutator : public MixedModeMutator {


Maybe with a TODO/issue then ?

apps/uma/_template/strategies.py

python/tvm/relay/backend/contrib/uma/api/codegen.py

src/relay/backend/contrib/uma/tir_to_runtime.cc

src/relay/backend/contrib/uma/targets.cc

manupak · 2022-08-05T08:46:45Z

replace AOTTestRunner by project API @manupa-arm

I am happy with AOTTestRunner as we envision that example/test to be useful for a embedded dev who wants to integrate few models to the their own embedded project. Also IIRC, this was @areusch suggestion :)

In same time, it might be also valuable to add a Project API for users that quickly want spin up a project to test the inference.
So they both have value and their place.

Therefore, I would say not "replace" but maybe just extend/add ? -- Also happy to push it to another PR

…line due to missing CLANG for i386

…at TE functions

areusch · 2022-08-09T16:41:39Z

thanks @MichaelJKlaiber for all the hard work here!

MichaelJKlaiber mentioned this pull request Jul 14, 2022

[Tracking Issue] UMA: Universal Modular Accelerator Interface #11260

Closed

21 tasks

MichaelJKlaiber changed the title ~~Uma v1.0~~ UMA v1.0 Jul 14, 2022

MichaelJKlaiber changed the title ~~UMA v1.0~~ [UMA] UMA v1.0 Jul 14, 2022

cbalint13 requested changes Jul 15, 2022

View reviewed changes

CMakeLists.txt Outdated Show resolved Hide resolved

MichaelJKlaiber force-pushed the uma_v1.0 branch from 8b917d7 to abb2976 Compare July 18, 2022 08:43

cbalint13 approved these changes Jul 18, 2022

View reviewed changes

sunggg reviewed Jul 18, 2022

View reviewed changes

areusch reviewed Jul 20, 2022

View reviewed changes

kslavka reviewed Jul 20, 2022

View reviewed changes

python/tvm/relay/backend/contrib/uma/tutorial.md Outdated Show resolved Hide resolved

manupak requested changes Jul 20, 2022

View reviewed changes

lhutton1 reviewed Jul 20, 2022

View reviewed changes

MichaelJKlaiber mentioned this pull request Jul 27, 2022

Uma MichaelJKlaiber/tvm#13

Merged

MichaelJKlaiber added a commit to MichaelJKlaiber/tvm that referenced this pull request Jul 27, 2022

Merge pull request #13 from MichaelJKlaiber/uma

212df81

UMAv1.0 updates according to comments and suggestions from PR apache#12087

areusch reviewed Jul 27, 2022

View reviewed changes

MichaelJKlaiber force-pushed the uma_v1.0 branch 3 times, most recently from 31b6107 to cac3060 Compare August 2, 2022 20:34

areusch approved these changes Aug 4, 2022

View reviewed changes

areusch requested a review from manupak August 4, 2022 15:47

sunggg approved these changes Aug 4, 2022

View reviewed changes

tests/python/contrib/test_uma/test_uma_pipeline.py Show resolved Hide resolved

manupak requested changes Aug 5, 2022

View reviewed changes

MichaelJKlaiber and others added 24 commits August 9, 2022 10:59

[UMA] lint updates

69ebe59

[UMA] Test updates

f2480d9

[UMA] fixes according to comments from PR 12087 discussion

3dab3f7

[UMA] lint updates

fa02920

[UMA] moved UMA _template file to apps

54160fb

[UMA] lint

f9aa26e

[UMA] Remove exceptions when dispatching over targets

960fef6

[UMA] vanilla pattern update

734d265

[UMA] added mobilenet integration test

6f100b9

[UMA] clang lint

e941421

Remove tir to runtime

011d2f2

[UMA] Use sequential for UMA relay passes

4875ed2

Use comparison against BYOC flow in test_partition

ccb4f4c

[UMA] tutorial update: moved code blocks to RST

1797ae4

[UMA] tutorial update and lint fixes

b1f02e1

[UMA] removing UMA from i386 build, as there is a fail in the CI pipe…

0cbe625

…line due to missing CLANG for i386

[BYOC-DNNL] covered case for sum node without attr

af16ae2

[UMA] pylint

ab670d1

[UMA] pylint

ad61655

[UMA] aot fix

34a3b69

[UMA] Changes PR review

ac2379e

[UMA] cc lint

cd536cc

[UMA] cc lint

0418ad8

Use better function name for te_lowering and annotate current target …

0f0b1bf

…at TE functions

MichaelJKlaiber force-pushed the uma_v1.0 branch from 1878882 to 0f0b1bf Compare August 9, 2022 09:02

areusch merged commit 4c4fcee into apache:main Aug 9, 2022

AndrewZhaoLuo mentioned this pull request Oct 4, 2022

TVM v0.10.0.rc0 Release Candidate Notes #12979

Closed

[UMA] UMA v1.0 #12087

[UMA] UMA v1.0 #12087

Uh oh!

Conversation

MichaelJKlaiber commented Jul 14, 2022

Uh oh!

cbalint13 left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

cbalint13 commented Jul 15, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sunggg left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cbalint13 commented Jul 19, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sunggg commented Jul 19, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

areusch left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cbalint13 commented Jul 20, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

MichaelJKlaiber commented Jul 20, 2022

Uh oh!

manupak left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

lhutton1 left a comment

cbalint13 left a comment •

edited

Loading

cbalint13 commented Jul 15, 2022 •

edited

Loading

cbalint13 commented Jul 19, 2022 •

edited

Loading

sunggg commented Jul 19, 2022 •

edited

Loading

cbalint13 commented Jul 20, 2022 •

edited

Loading

manupak left a comment •

edited

Loading

MichaelJKlaiber commented Aug 3, 2022 •

edited

Loading