masahi edited a comment on pull request #7441:
URL: https://github.com/apache/tvm/pull/7441#issuecomment-777181989
Thanks, I was planning to work on unique next week, happy to collaborate.
I can add TIR unqiue impl both cpu and gpu later. We can add relay
boilarplate, temp impl in
masahi commented on pull request #7441:
URL: https://github.com/apache/tvm/pull/7441#issuecomment-777181989
Thanks, I was planning to work on unique next week, happy to collaborate.
I can add TIR unqiue impl both cpu and gpu later.
codeislife99 opened a new pull request #7442:
URL: https://github.com/apache/tvm/pull/7442
-
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above
codeislife99 commented on a change in pull request #7441:
URL: https://github.com/apache/tvm/pull/7441#discussion_r574221285
##
File path: src/relay/op/algorithm/unique.cc
##
@@ -0,0 +1,147 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more
codeislife99 commented on a change in pull request #7441:
URL: https://github.com/apache/tvm/pull/7441#discussion_r574220931
##
File path: python/tvm/relay/op/strategy/generic.py
##
@@ -1395,3 +1395,28 @@ def cumsum_strategy(attrs, inputs, out_type, target):
codeislife99 commented on a change in pull request #7441:
URL: https://github.com/apache/tvm/pull/7441#discussion_r574219984
##
File path: python/tvm/relay/op/_transform.py
##
@@ -885,3 +886,29 @@ def where_shape_func(attrs, inputs, _):
out_shape =
ymwangg opened a new pull request #7441:
URL: https://github.com/apache/tvm/pull/7441
This PR adds the tensorflow `unique` operator as described in
https://www.tensorflow.org/api_docs/python/tf/unique.
I'm not sure I follow the best practices. Comments and suggestions are
welcome.
electriclilies opened a new pull request #7440:
URL: https://github.com/apache/tvm/pull/7440
This PR adds max mode to ROI align.
The ONNX version of max mode in ROI align is incorrect (see
https://github.com/microsoft/onnxruntime/issues/6146), so I'm not testing ROI
align with max
junrushao1994 commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777100312
I agree with @leandron and @comaniac. Perhaps a good way is to have both
`list_kinds` and `list_tags` correspondingly
junrushao1994 edited a comment on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777098967
Thanks for the discussion, and sorry for the late reply!
To clarify:
- Target `tag`s are an alias and an efficient way to create concrete
targets. For
junrushao1994 commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777098967
Thanks for the discussion, and sorry for the late reply!
To clarify:
- Target `tag`s are an alias and an efficient way to create concrete
targets. For example,
merrymercy edited a comment on pull request #7439:
URL: https://github.com/apache/tvm/pull/7439#issuecomment-777086490
We also have to update the commands in
`tvm/tutorials/auto_scheduler/tune_network_*.py`
merrymercy commented on pull request #7439:
URL: https://github.com/apache/tvm/pull/7439#issuecomment-777086490
We also have to update the command in
`tvm/tutorials/auto_scheduler/tune_network*.py`
comaniac opened a new pull request #7439:
URL: https://github.com/apache/tvm/pull/7439
The distill record in auto_scheduler is broken due to the recent change
(#7344). This PR fixes the function and make it self-contained.
cc @merrymercy
tqchen commented on pull request #7437:
URL: https://github.com/apache/tvm/pull/7437#issuecomment-777076046
Please apply the clang format via `./tests/lint/git-clang-format.sh -i
upstream/main`
This is an automated message
This is an automated email from the ASF dual-hosted git repository.
junrushao pushed a commit to branch ci-docker-staging
in repository https://gitbox.apache.org/repos/asf/tvm.git
The following commit(s) were added to refs/heads/ci-docker-staging by this push:
new db93cb8 retrigger CI
comaniac edited a comment on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777062402
IMHO, it would be better to use a more consistant name. i.e., `list_kinds`.
This also aligns to the naming convention of `list_tags`.
zxybazh commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777065443
Oh, I see. It's following the `list_tags` function in `tag.py`. Sounds good
to me.
This is an automated message
leandron commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777065155
> On Python-side function, I think `names` is a bit vague. Instead, `kinds`
is more clear and consistent with the other `libs` function name.
I think @comaniac makes a good
comaniac commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777062402
IMHO, it would be better to use a more consistant name. i.e., `list_kinds`.
This also aligns to the naming convension of `list_tags`.
zxybazh edited a comment on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777061935
On Python-side function, I think `names` is a bit vague. Instead, `kinds` is
more clear and consistent with the other `libs` function name.
zxybazh commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777061935
On Python-side function, I think `names` is a bit vague. I think `kinds` is
more clear and consistent with the `libs` function name.
leandron edited a comment on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777055150
Thanks for the replies @comaniac @zxybazh. One final thing from me (to get
you on the right timezone) - just so that we update with a potential final
version of the PR.
leandron edited a comment on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777055150
Thanks for the replies @comaniac @zxybazh. One final thing from me (to get
you on the right timezone) - just so that we update gwith a potential final
version of the PR.
leandron commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777055150
Thanks for the replies @comaniac @zxybazh. One final thing (to get you on
the right timezone) - just so that we update gwith a potential final version of
the PR.
Any
mbrookhart commented on pull request #7438:
URL: https://github.com/apache/tvm/pull/7438#issuecomment-777047550
Definitely a needed improvement, we'll see what it uncovers.
This is an automated message from the Apache
electriclilies opened a new pull request #7438:
URL: https://github.com/apache/tvm/pull/7438
In the onnx test, we were setting target and ctx to llvm and cpu every time
we ran the TVM version of the op, regardless of the target and ctx actually
passed in. The result is that the accuracy
comaniac commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777041915
Agree on the ListTargetKinds in this case. ListTargetKindNames doesn't make
it more clear to me.
This is an
gromero commented on pull request #7333:
URL: https://github.com/apache/tvm/pull/7333#issuecomment-777040897
> @gromero yes, I think that's roughly what GH does when someone hits merge.
Got it :+1:
> the review comments do stick around, but the context is usually
zxybazh commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777034419
I agree. For the function name selection, I would prefer to use
`ListTargetKind(s)` as @junrushao1994 suggested, somehow more consistent with
the meaning of `kind`. `Names` still
areusch commented on pull request #7333:
URL: https://github.com/apache/tvm/pull/7333#issuecomment-777030634
@gromero yes, I think that's roughly what GH does when someone hits merge.
the review comments do stick around, but the context is usually obliterated,
which is usually
leandron commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777029852
> I think the PR is just to demonstrate the supported backend types, which
is called `kind`, as @comaniac mentioned. Though accessible via code, a
function would be helpful. `tag`
zxybazh commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-777022644
I think the PR is just to demonstrate the supported backend types, which is
called `kind`, as @comaniac mentioned. Though accessible via code, a function
would be helpful. `tag` is
trevor-m commented on a change in pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#discussion_r574065135
##
File path: cmake/modules/contrib/TensorRT.cmake
##
@@ -28,7 +28,9 @@ if(USE_TENSORRT_CODEGEN)
file(GLOB RUNTIME_TENSORRT_SRCS
vinx13 opened a new pull request #7437:
URL: https://github.com/apache/tvm/pull/7437
Return original expression if the expression can't be resolved. Otherwise,
passing IterMapExpr to Simplify causes error.
@tqchen
trevor-m commented on a change in pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#discussion_r574065135
##
File path: cmake/modules/contrib/TensorRT.cmake
##
@@ -28,7 +28,9 @@ if(USE_TENSORRT_CODEGEN)
file(GLOB RUNTIME_TENSORRT_SRCS
This is an automated email from the ASF dual-hosted git repository.
wuwei pushed a commit to branch fix/iter_map
in repository https://gitbox.apache.org/repos/asf/tvm.git
commit 693aa52c733c94f82b8c3f75a1bac245aca164ce
Author: Wuwei Lin
AuthorDate: Wed Feb 10 12:41:17 2021 -0800
[Arith]
This is an automated email from the ASF dual-hosted git repository.
wuwei pushed a change to branch fix/iter_map
in repository https://gitbox.apache.org/repos/asf/tvm.git.
at 693aa52 [Arith] Fix iter_affine_map with non-const extent
This branch includes the following new commits:
gromero edited a comment on pull request #7333:
URL: https://github.com/apache/tvm/pull/7333#issuecomment-777008938
> @gromero easiest way to patch given merges might be something like:
>
> ```
> $ git fetch origin main
> $ git fetch areusch
> $ git diff
gromero commented on pull request #7333:
URL: https://github.com/apache/tvm/pull/7333#issuecomment-777008938
> @gromero easiest way to patch given merges might be something like:
>
> ```
> $ git fetch origin main
> $ git fetch areusch
> $ git diff areusch/...origin/main
tkonolige commented on a change in pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#discussion_r574054099
##
File path: cmake/modules/contrib/TensorRT.cmake
##
@@ -28,7 +28,9 @@ if(USE_TENSORRT_CODEGEN)
file(GLOB RUNTIME_TENSORRT_SRCS
trevor-m commented on a change in pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#discussion_r574051855
##
File path: cmake/modules/contrib/TensorRT.cmake
##
@@ -28,7 +28,9 @@ if(USE_TENSORRT_CODEGEN)
file(GLOB RUNTIME_TENSORRT_SRCS
trevor-m commented on a change in pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#discussion_r574051332
##
File path: cmake/modules/contrib/TensorRT.cmake
##
@@ -28,7 +28,9 @@ if(USE_TENSORRT_CODEGEN)
file(GLOB RUNTIME_TENSORRT_SRCS
tkonolige commented on a change in pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#discussion_r574045801
##
File path: cmake/modules/contrib/TensorRT.cmake
##
@@ -28,7 +28,9 @@ if(USE_TENSORRT_CODEGEN)
file(GLOB RUNTIME_TENSORRT_SRCS
This is an automated email from the ASF dual-hosted git repository.
wuwei pushed a change to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git.
from 9175c6d TRT Dynamic Reshape Fix (#7412)
add 12c6b70 Simplify full broadcast (#7423)
No new revisions were added by
vinx13 commented on pull request #7423:
URL: https://github.com/apache/tvm/pull/7423#issuecomment-776974366
Thanks @mbrookhart @masahi
This is an automated message from the Apache Git Service.
To respond to the message,
vinx13 merged pull request #7423:
URL: https://github.com/apache/tvm/pull/7423
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the
comaniac commented on pull request #7412:
URL: https://github.com/apache/tvm/pull/7412#issuecomment-776972346
Thanks @codeislife99 @trevor-m
This is an automated message from the Apache Git Service.
To respond to the
This is an automated email from the ASF dual-hosted git repository.
comaniac pushed a change to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git.
from 3863e09 [FIX,CMAKE] Only set Clang flags for C++ files (#7424)
add 9175c6d TRT Dynamic Reshape Fix (#7412)
No
comaniac merged pull request #7412:
URL: https://github.com/apache/tvm/pull/7412
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to
codeislife99 commented on pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#issuecomment-776972280
Thanks Trevor.
This is an automated message from the Apache Git Service.
To respond to the message, please
areusch commented on a change in pull request #7333:
URL: https://github.com/apache/tvm/pull/7333#discussion_r574006510
##
File path: python/tvm/micro/build.py
##
@@ -55,15 +57,62 @@ def path(self):
CRT_RUNTIME_LIB_NAMES = ["utvm_rpc_server", "utvm_rpc_common", "common"]
tqchen commented on issue #7246:
URL: https://github.com/apache/tvm/issues/7246#issuecomment-776946615
ping @zhuwenxi please let us know about the status and anything we can help
This is an automated message from the Apache
comaniac commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-776941068
Ah I messed up with `tag.cc` and `target_kind.cc`...
So basically `kind` refers to the backend, such as `llvm`, `cpu`, `gpu`,
`cuda`, and `tag` refers to a complete target
leandron commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-776924911
> I reviewed the changes again based on the discussion. Why don't we just
add a syntactic sugar`Target.list_tags().keys()` in the Python side so that we
don't need a new function?
areusch commented on pull request #7333:
URL: https://github.com/apache/tvm/pull/7333#issuecomment-776911131
@gromero easiest way to patch given merges might be something like:
```
$ git fetch origin main
$ git fetch areusch
$ git diff areusch/...origin/main >patch # NOTE 3
leandron commented on pull request #7422:
URL: https://github.com/apache/tvm/pull/7422#issuecomment-776910235
Yeah, fair enough. I guess we can merge this @tqchen, maybe?
This is an automated message from the Apache Git
trevor-m commented on a change in pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#discussion_r573956757
##
File path: cmake/modules/contrib/TensorRT.cmake
##
@@ -28,7 +28,9 @@ if(USE_TENSORRT_CODEGEN)
file(GLOB RUNTIME_TENSORRT_SRCS
trevor-m commented on a change in pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#discussion_r573956757
##
File path: cmake/modules/contrib/TensorRT.cmake
##
@@ -28,7 +28,9 @@ if(USE_TENSORRT_CODEGEN)
file(GLOB RUNTIME_TENSORRT_SRCS
trevor-m commented on a change in pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#discussion_r573956757
##
File path: cmake/modules/contrib/TensorRT.cmake
##
@@ -28,7 +28,9 @@ if(USE_TENSORRT_CODEGEN)
file(GLOB RUNTIME_TENSORRT_SRCS
comaniac commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-776891969
I reviewed the changes again based on the discussion. Why don't we just add
a syntactic sugar`Target.list_tags().keys()` in the Python side so that we
don't need a new function?
tkonolige commented on a change in pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#discussion_r573937245
##
File path: cmake/modules/contrib/TensorRT.cmake
##
@@ -28,7 +28,9 @@ if(USE_TENSORRT_CODEGEN)
file(GLOB RUNTIME_TENSORRT_SRCS
comaniac commented on pull request #7436:
URL: https://github.com/apache/tvm/pull/7436#issuecomment-776882300
Also cc @tkonolige
This is an automated message from the Apache Git Service.
To respond to the message, please
trevor-m opened a new pull request #7436:
URL: https://github.com/apache/tvm/pull/7436
After https://github.com/apache/tvm/pull/7417 the cmake rules for TensorRT
and ACL BYOC would cause the runtime sources to be compiled twice if both
runtime and codegen are enabled together, giving the
jtuyls commented on pull request #7422:
URL: https://github.com/apache/tvm/pull/7422#issuecomment-776876505
@leandron The motivation is that I have been trying out TVM with ROCm and it
took me a bit of time to put the necessary pieces together so I think this
improves usability and might
euntaik commented on a change in pull request #7400:
URL: https://github.com/apache/tvm/pull/7400#discussion_r573867907
##
File path: python/tvm/driver/tvmc/frontends.py
##
@@ -241,43 +241,10 @@ def load(self, path, shape_dict=None):
if version != 3:
leandron commented on a change in pull request #7400:
URL: https://github.com/apache/tvm/pull/7400#discussion_r573844512
##
File path: python/tvm/driver/tvmc/frontends.py
##
@@ -241,43 +241,10 @@ def load(self, path, shape_dict=None):
if version != 3:
euntaik commented on a change in pull request #7400:
URL: https://github.com/apache/tvm/pull/7400#discussion_r573843547
##
File path: python/tvm/driver/tvmc/frontends.py
##
@@ -241,43 +241,10 @@ def load(self, path, shape_dict=None):
if version != 3:
euntaik commented on a change in pull request #7400:
URL: https://github.com/apache/tvm/pull/7400#discussion_r573843007
##
File path: python/tvm/relay/frontend/tflite.py
##
@@ -3539,7 +3539,45 @@ def get_tensor_name(subgraph, tensor_idx):
return
ANSHUMAN87 opened a new pull request #7435:
URL: https://github.com/apache/tvm/pull/7435
1: Sparse Add Op added in Topi.
2: TF Frontend support added.
This is an automated message from the Apache Git Service.
To
gromero commented on pull request #7333:
URL: https://github.com/apache/tvm/pull/7333#issuecomment-776787476
> @gromero thanks for the review. unfortunately if you force-push (I.e.
rebase) it loses the context around the previous comments, and I like that less
:/.
@areusch yeah..
leandron commented on a change in pull request #7400:
URL: https://github.com/apache/tvm/pull/7400#discussion_r573808458
##
File path: python/tvm/relay/frontend/tflite.py
##
@@ -3539,7 +3539,45 @@ def get_tensor_name(subgraph, tensor_idx):
return
This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a change to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git.
from 68b7e71 [Relay][Topi][CPU] Dense with weight transform (#7404)
add 3863e09 [FIX,CMAKE] Only set Clang flags for
tqchen merged pull request #7424:
URL: https://github.com/apache/tvm/pull/7424
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the
masahi edited a comment on pull request #7425:
URL: https://github.com/apache/tvm/pull/7425#issuecomment-776703600
@tqchen @junrushao1994 @vinx13
I went through the passes and here is my summary:
* `VectorizeLoop`: Need to disallow a while loop inside a vectorized loop.
Without
masahi edited a comment on pull request #7425:
URL: https://github.com/apache/tvm/pull/7425#issuecomment-776703600
@tqchen @junrushao1994 @vinx13
I went through passes and here is my summary:
* `VectorizeLoop`: Need to disallow a while loop inside a vectorized loop.
Without it,
masahi edited a comment on pull request #7425:
URL: https://github.com/apache/tvm/pull/7425#issuecomment-776703600
@tqchen @junrushao1994 @vinx13
I went through passes and here is my summary:
* `VectorizeLoop`: Need to disallow a while loop inside a vectorized loop.
Without it,
masahi edited a comment on pull request #7425:
URL: https://github.com/apache/tvm/pull/7425#issuecomment-776703600
@tqchen @junrushao1994 @vinx13
I went through passes and here is my summary:
* `VectorizeLoop`: Need to disallow a while loop inside a vectorized loop.
Without it,
masahi commented on pull request #7425:
URL: https://github.com/apache/tvm/pull/7425#issuecomment-776703600
@tqchen @junrushao1994 @vinx13
I went through passes and here is my summary:
* `VectorizeLoop`: Need to disallow a while loop inside a vectorized loop.
Without it, no
manupa-arm commented on a change in pull request #7333:
URL: https://github.com/apache/tvm/pull/7333#discussion_r573597149
##
File path: python/tvm/micro/build.py
##
@@ -55,15 +57,62 @@ def path(self):
CRT_RUNTIME_LIB_NAMES = ["utvm_rpc_server", "utvm_rpc_common", "common"]
mbaret commented on pull request #7427:
URL: https://github.com/apache/tvm/pull/7427#issuecomment-776610095
> Therefore, to keep things consistent, do you think we should be more
specific here and say like `ListTargetKind` instead of using `names`?
I would be slightly concerned that
euntaik commented on a change in pull request #7400:
URL: https://github.com/apache/tvm/pull/7400#discussion_r573581898
##
File path: python/tvm/relay/frontend/tflite.py
##
@@ -3588,8 +3643,14 @@ def from_tflite(model, shape_dict, dtype_dict):
exp_tab = ExprTable()
leandron commented on a change in pull request #7400:
URL: https://github.com/apache/tvm/pull/7400#discussion_r573551305
##
File path: python/tvm/relay/frontend/tflite.py
##
@@ -3588,8 +3643,14 @@ def from_tflite(model, shape_dict, dtype_dict):
exp_tab = ExprTable()
euntaik commented on a change in pull request #7400:
URL: https://github.com/apache/tvm/pull/7400#discussion_r573533135
##
File path: python/tvm/relay/frontend/tflite.py
##
@@ -3588,8 +3643,14 @@ def from_tflite(model, shape_dict, dtype_dict):
exp_tab = ExprTable()
euntaik commented on a change in pull request #7400:
URL: https://github.com/apache/tvm/pull/7400#discussion_r573533135
##
File path: python/tvm/relay/frontend/tflite.py
##
@@ -3588,8 +3643,14 @@ def from_tflite(model, shape_dict, dtype_dict):
exp_tab = ExprTable()
85 matches
Mail list logo