mbaret commented on pull request #5409:
URL: https://github.com/apache/incubator-tvm/pull/5409#issuecomment-621253051
I've implemented an alternative fix in partition_graph.cc by explicitly
binding constant tuples (see
[5476](https://github.com/apache/incubator-tvm/pull/5476)). This means
roastduck edited a comment on pull request #5465:
URL: https://github.com/apache/incubator-tvm/pull/5465#issuecomment-621285171
Made the pass default when lowering. So that not only generated device code,
but also lowered IR in simple mode, will reflect this pass.
roastduck commented on pull request #5465:
URL: https://github.com/apache/incubator-tvm/pull/5465#issuecomment-621285171
Made the pass default when lowering.
This is an automated message from the Apache Git Service.
To
tqchen edited a comment on pull request #5465:
URL: https://github.com/apache/incubator-tvm/pull/5465#issuecomment-621242357
Thanks @roastduck
Given that it is a quit fundamental construct that can leads to concise
generated code(e.g. ? in CUDA), I think we should keep it
mbaret opened a new pull request #5476:
URL: https://github.com/apache/incubator-tvm/pull/5476
Constant tuples were not being propagated into partitions making it
impossible to codegen functions like concatenate. This extends the constant
binding that happens in PartitionGraph to also
tqchen commented on pull request #5417:
URL: https://github.com/apache/incubator-tvm/pull/5417#issuecomment-621237313
@weberlo @tmoreau89 @liangfu @u99127 please take another look and
http://tvm.apache.org/docs/contribute/code_review.html#approve-and-request-changes-explicitly
tqchen commented on pull request #5465:
URL: https://github.com/apache/incubator-tvm/pull/5465#issuecomment-621242357
Thanks @roastduck if the goal is mainly to utilize LoopParition, perhaps a
better way would be to insert likely into the condition and enable
LoopPartition to deal with
mbaret commented on pull request #5473:
URL: https://github.com/apache/incubator-tvm/pull/5473#issuecomment-621250794
cc @anijain2305
This is an automated message from the Apache Git Service.
To respond to the message,
This is an automated email from the ASF dual-hosted git repository.
moreau pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm-vta.git
The following commit(s) were added to refs/heads/master by this push:
new 22e4b88 [Hardware][Xilinx] utilize
tqchen commented on issue #5464:
URL: https://github.com/apache/incubator-tvm/issues/5464#issuecomment-621240765
cc @kazum @masahi
This is an automated message from the Apache Git Service.
To respond to the message, please
tqchen commented on issue #5446:
URL: https://github.com/apache/incubator-tvm/issues/5446#issuecomment-621241256
Thanks for start the discussion, Please open a new discussion thread on
https://discuss.tvm.ai/
This is an
mbaret commented on a change in pull request #5474:
URL: https://github.com/apache/incubator-tvm/pull/5474#discussion_r417383328
##
File path: tests/python/frontend/tflite/test_forward.py
##
@@ -1896,6 +1896,41 @@ def test_forward_mediapipe_hand_landmark():
anijain2305 commented on pull request #5473:
URL: https://github.com/apache/incubator-tvm/pull/5473#issuecomment-621313035
Thanks @mbaret This is merged!
This is an automated message from the Apache Git Service.
To respond
This is an automated email from the ASF dual-hosted git repository.
anijain2305 pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git
The following commit(s) were added to refs/heads/master by this push:
new 95a816c [TFLITE] Match TFLite
roastduck commented on pull request #5465:
URL: https://github.com/apache/incubator-tvm/pull/5465#issuecomment-621282600
> Given that it is a quit fundamental construct that can leads to concise
generated code(e.g. ? in CUDA), I think we should keep it throughout the
lowering process by
tqchen commented on issue #4102:
URL: https://github.com/apache/incubator-tvm/issues/4102#issuecomment-621363297
NOTE: the Mainline CI now uses tensorflow 2.0
This is an automated message from the Apache Git Service.
To
This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a change to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git.
from f1e87f1 [CODEGEN][CUDA] Fix a bug when vectorized load was
involved for… (#5428)
add 4d8816c [CI]
tqchen commented on pull request #5436:
URL: https://github.com/apache/incubator-tvm/pull/5436#issuecomment-621364933
NOTE: the ci-cpu is still blocked by
https://github.com/apache/incubator-tvm/issues/5455 ci-gpu is now updated. If
we can update the testcase to pre-check the env before
icemelon9 commented on pull request #5357:
URL: https://github.com/apache/incubator-tvm/pull/5357#issuecomment-621364453
@Menooker Could you share the script that can reproduce the error? It's a
bit weird that `_alter_conv2d_layout` receive a conv2d op with `NCHWc` layout.
mbaret commented on pull request #5360:
URL: https://github.com/apache/incubator-tvm/pull/5360#issuecomment-621346205
@anijain2305 @zhiics could you take a look at the changes I've made and see
if this is now OK?
This is an
This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git
The following commit(s) were added to refs/heads/master by this push:
new f1e87f1 [CODEGEN][CUDA] Fix a bug when
tqchen commented on pull request #5428:
URL: https://github.com/apache/incubator-tvm/pull/5428#issuecomment-621350989
Thanks @wpan11nv @boh-inspur !
This is an automated message from the Apache Git Service.
To respond to
huajsj commented on pull request #7:
URL: https://github.com/apache/incubator-tvm-vta/pull/7#issuecomment-621379576
@zhanghaohit, thanks for the contribution, do we know the root cause for
this early return issue? a sleep solution may not always work and would impact
performance, we
tqchen edited a comment on pull request #5436:
URL: https://github.com/apache/incubator-tvm/pull/5436#issuecomment-621364933
NOTE:
- the ci-cpu is still blocked by
https://github.com/apache/incubator-tvm/issues/5455
- but ci-gpu is now updated to sync with the latest state. If we
This is an automated email from the ASF dual-hosted git repository.
zhic pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git
The following commit(s) were added to refs/heads/master by this push:
new 046b0d9 [BYOC] Bind constant tuples in
icemelon9 commented on a change in pull request #5357:
URL: https://github.com/apache/incubator-tvm/pull/5357#discussion_r417472667
##
File path: python/tvm/relay/op/strategy/x86.py
##
@@ -84,8 +88,13 @@ def conv2d_strategy_cpu(attrs, inputs, out_type, target):
raise
tqchen commented on pull request #5392:
URL: https://github.com/apache/incubator-tvm/pull/5392#issuecomment-621363701
NOTE: the mainline ci-gpu now takes in effect
This is an automated message from the Apache Git Service.
To
kevinthesun commented on a change in pull request #4312:
URL: https://github.com/apache/incubator-tvm/pull/4312#discussion_r417499071
##
File path: python/tvm/relay/frontend/tensorflow.py
##
@@ -612,6 +615,51 @@ def _impl(inputs, attr, params, mod):
out =
anijain2305 commented on a change in pull request #5439:
URL: https://github.com/apache/incubator-tvm/pull/5439#discussion_r417583390
##
File path: tests/python/frontend/tflite/test_forward.py
##
@@ -1933,21 +1937,41 @@ def test_forward_qnn_mobilenet_v3_net():
# SSD Mobilenet
zhiics edited a comment on pull request #5457:
URL: https://github.com/apache/incubator-tvm/pull/5457#issuecomment-621515297
@MarisaKirisame Sorry for the late response. Yeah, transforming it to ANF
should work as well and seems like better fix. Let me try it.
BTW, this would also
kevinthesun commented on a change in pull request #5429:
URL: https://github.com/apache/incubator-tvm/pull/5429#discussion_r417650019
##
File path: python/tvm/relay/_parser.py
##
@@ -114,7 +114,10 @@ def convert(self, v):
def __call__(self, args, attrs, type_args):
kevinthesun commented on pull request #5457:
URL: https://github.com/apache/incubator-tvm/pull/5457#issuecomment-621505667
@MarisaKirisame The current use cases which can trigger is this issue are
mainly from TensorFlow OD models. For those models, constant tensors which can
be feed into
yongwww commented on a change in pull request #5429:
URL: https://github.com/apache/incubator-tvm/pull/5429#discussion_r417638357
##
File path: python/tvm/relay/_parser.py
##
@@ -114,7 +114,10 @@ def convert(self, v):
def __call__(self, args, attrs, type_args):
anijain2305 opened a new pull request #5477:
URL: https://github.com/apache/incubator-tvm/pull/5477
New test will follow up
This is an automated message from the Apache Git Service.
To respond to the message, please log on
zhiics commented on pull request #5457:
URL: https://github.com/apache/incubator-tvm/pull/5457#issuecomment-621515297
@MarisaKirisame Sorry for the late response. Yeah, transforming it to ANF
should work as well and seems like better fix. Let me try it.
tqchen opened a new pull request #5478:
URL: https://github.com/apache/incubator-tvm/pull/5478
Allows Optional and module to be passed with the right type code.
This is an automated message from the Apache Git
tqchen commented on pull request #5478:
URL: https://github.com/apache/incubator-tvm/pull/5478#issuecomment-621527940
cc @zhiics @ZihengJiang @yzhliu @jroesch @junrushao1994
This is an automated message from the Apache Git
anijain2305 opened a new pull request #5479:
URL: https://github.com/apache/incubator-tvm/pull/5479
TBD
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use
remotego commented on pull request #7:
URL: https://github.com/apache/incubator-tvm-vta/pull/7#issuecomment-621581807
Thanks @huajsj @tmoreau89 for pointing out the potential performance issue
for this fix.
The root cause of the issue is because on line 124, we instruct the VTA to
tmoreau89 commented on a change in pull request #7:
URL: https://github.com/apache/incubator-tvm-vta/pull/7#discussion_r417732849
##
File path: src/pynq/pynq_driver.cc
##
@@ -126,6 +127,10 @@ class VTADevice {
VTAWriteMappedReg(vta_compute_handle_, 0x0, VTA_AUTORESTART);
tmoreau89 commented on pull request #7:
URL: https://github.com/apache/incubator-tvm-vta/pull/7#issuecomment-621590481
Thank you @remotego for this logical explanation. The fix makes sense, and
is highly likely not to affect performance negatively.
spectrometerHBH opened a new pull request #5483:
URL: https://github.com/apache/incubator-tvm/pull/5483
## Several Points
1. Rename vars/size_vars/buffers to use name hints to identify them
2. A node is either in META or not. If ref A is put in META, ref B is
contained in A, then
tmoreau89 commented on pull request #8:
URL: https://github.com/apache/incubator-tvm-vta/pull/8#issuecomment-621597493
Thank you for the explanation. Do you mind breaking down what happens in the
32x32 configuration that causes correctness to fail (since the dependence
analysis waive is
tmoreau89 commented on pull request #8:
URL: https://github.com/apache/incubator-tvm-vta/pull/8#issuecomment-621599822
I'll think some more about this problem overnight, thanks for bringing
attention to this issue @zhanghaohit @remotego
tmoreau89 commented on pull request #8:
URL: https://github.com/apache/incubator-tvm-vta/pull/8#issuecomment-621599637
Perhaps one way to look at this is to start with DISTANCE=3 by default. And
if the II>1 for the GEMM, we issue a warning telling the user to increase the
distance to 4.
tqchen commented on pull request #5483:
URL: https://github.com/apache/incubator-tvm/pull/5483#issuecomment-621600022
cc @jroesch @antinucleon @junrushao1994 @Hzfengsy @mbrookhart
This is an automated message from the
huajsj commented on pull request #7:
URL: https://github.com/apache/incubator-tvm-vta/pull/7#issuecomment-621612870
@remotego, Thanks for the detailed explain, but for this problem I think you
may have a glance on related pynq driver logic, actually there is no "racing
condition" between
anijain2305 commented on pull request #5477:
URL: https://github.com/apache/incubator-tvm/pull/5477#issuecomment-621543411
@tqchen This PR is for the web-data PR -
https://github.com/dmlc/web-data/pull/247
This is an
tmoreau89 commented on pull request #5:
URL: https://github.com/apache/incubator-tvm-vta/pull/5#issuecomment-621552938
Very exciting development @huajsj ! I'm looking forward to the C++ example!
This is an automated message
tmoreau89 commented on pull request #7:
URL: https://github.com/apache/incubator-tvm-vta/pull/7#issuecomment-621552736
Thanks @zhanghaohit for the fix. I'm curious when the early completion
occurs as @huajsj pointed out.
Ideally the driver shouldn't rely on polling and instead use
zhanghaohit commented on pull request #8:
URL: https://github.com/apache/incubator-tvm-vta/pull/8#issuecomment-621567733
> Thanks @zhanghaohit, for catching this error. I agree with the fact that
the dependence should not be hardcoded. However in order to not to add too many
parameters in
yzhliu opened a new pull request #5482:
URL: https://github.com/apache/incubator-tvm/pull/5482
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL
tmoreau89 commented on pull request #8:
URL: https://github.com/apache/incubator-tvm-vta/pull/8#issuecomment-621598201
@remotego I just read your comment that must have been posted as I wrote my
reply. Your explanation makes sense, ultimately the critical path in the
accumulation is READ
This is an automated email from the ASF dual-hosted git repository.
moreau pushed a change to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git.
from 046b0d9 [BYOC] Bind constant tuples in graph partitioner (#5476)
add 684f2d7 [VTA][Runtime] clear sram
zhanghaohit commented on pull request #5471:
URL: https://github.com/apache/incubator-tvm/pull/5471#issuecomment-621572608
> Thanks @zhanghaohit for the proposed change, moving away from a hardcoded
value is the right move!
>
> I find the `ACC_DEP_DISTANCE` to be confusing however.
remotego edited a comment on pull request #7:
URL: https://github.com/apache/incubator-tvm-vta/pull/7#issuecomment-621581807
Thanks @huajsj @tmoreau89 for pointing out the potential performance issue
for this fix.
The root cause of the issue is because on line 124, we instruct the
remotego commented on pull request #8:
URL: https://github.com/apache/incubator-tvm-vta/pull/8#issuecomment-621593259
> Thanks @zhanghaohit, for catching this error. I agree with the fact that
the dependence should not be hardcoded. However in order to not to add too many
parameters in
tmoreau89 commented on pull request #8:
URL: https://github.com/apache/incubator-tvm-vta/pull/8#issuecomment-621598572
I assume that if you have an FPGA that needs to be clocked at a high
frequency, you'll likely end up with more cycles.
remotego commented on pull request #8:
URL: https://github.com/apache/incubator-tvm-vta/pull/8#issuecomment-621601302
> Perhaps one way to look at this is to start with DISTANCE=3 by default.
And if the II>1 for the GEMM, we issue a warning telling the user to increase
the distance to 4.
Menooker commented on pull request #5357:
URL: https://github.com/apache/incubator-tvm/pull/5357#issuecomment-621567041
> @Menooker Could you share the script that can reproduce the error? It's a
bit weird that `_alter_conv2d_layout` receive a conv2d op with `NCHWc` layout.
My
huajsj opened a new pull request #5481:
URL: https://github.com/apache/incubator-tvm/pull/5481
Issue1:
When doing vta compile in xilinx FPGA, the pynqdriver.cc report can
not find
Issue2:
Run vta on pynq board , libvta.so load failed.
solution:
fixed VTA.make
tobegit3hub commented on issue #4464:
URL: https://github.com/apache/incubator-tvm/issues/4464#issuecomment-621593883
The PR has been merged and we will close this issue.
This is an automated message from the Apache Git
tqchen opened a new pull request #5484:
URL: https://github.com/apache/incubator-tvm/pull/5484
This PR refactors the RPC protocol to make it more modularized.
- RPCSession: represent a set of features that need to be implemented
- RPCEndPont: End point that forwards the RPCSession
tqchen commented on pull request #5484:
URL: https://github.com/apache/incubator-tvm/pull/5484#issuecomment-621598079
cc @FrozenGene @merrymercy @eqy @ZihengJiang @areusch @ajtulloch
@antinucleon @junrushao1994 @jroesch @kazum @yzhliu
tqchen commented on pull request #5484:
URL: https://github.com/apache/incubator-tvm/pull/5484#issuecomment-621597882
RFC https://discuss.tvm.ai/t/modularize-rpc-infra/6544
This is an automated message from the Apache Git
This is an automated email from the ASF dual-hosted git repository.
ziheng pushed a change to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git.
from 684f2d7 [VTA][Runtime] clear sram index in reset (#5470)
add 90b08f5 [intrin] a few more math
tmoreau89 commented on pull request #8:
URL: https://github.com/apache/incubator-tvm-vta/pull/8#issuecomment-621551928
Thanks @zhanghaohit, for catching this error. I agree with the fact that the
dependence should not be hardcoded. However in order to not to add too many
parameters in the
siju-samuel opened a new pull request #5480:
URL: https://github.com/apache/incubator-tvm/pull/5480
@anijain2305 @masahi please help to review this PR.
Thanks.
This is an automated message from the Apache Git Service.
srkreddy1238 commented on issue #4102:
URL: https://github.com/apache/incubator-tvm/issues/4102#issuecomment-621595847
Good to see CI moving to 2.0.
On this context.
- I have upgraded TF test cases to use pure 2.0 API (with out compat v1,
tf.graph, tf.session ...etc.).
-
tmoreau89 commented on pull request #8:
URL: https://github.com/apache/incubator-tvm-vta/pull/8#issuecomment-621599402
So to conclude, I agree with you that we may want this parameter to be
larger than 2 cycles if the hardware pipeline becomes more complicated.
So the big question
zhiics commented on a change in pull request #5360:
URL: https://github.com/apache/incubator-tvm/pull/5360#discussion_r417496245
##
File path: python/tvm/relay/qnn/op/qnn.py
##
@@ -180,15 +180,16 @@ def concatenate(data,
The concatenated quantized tensor.
"""
-
siju-samuel opened a new pull request #5475:
URL: https://github.com/apache/incubator-tvm/pull/5475
Supported Logical Not for TFLite frontend.
@FrozenGene @masahi Please help to review and merge this PR.
TIA
This
roastduck commented on pull request #5465:
URL: https://github.com/apache/incubator-tvm/pull/5465#issuecomment-621159263
Fixed the breaking `fifo_buffer` test by avoiding passing the same buffer as
different arguments to a CUDA kernel.
roastduck commented on a change in pull request #5345:
URL: https://github.com/apache/incubator-tvm/pull/5345#discussion_r417265081
##
File path: python/tvm/relay/qnn/op/legalizations.py
##
@@ -20,8 +20,8 @@
import tvm
from tvm import relay
+import numpy as np
Review
Menooker commented on a change in pull request #5357:
URL: https://github.com/apache/incubator-tvm/pull/5357#discussion_r417267279
##
File path: python/tvm/relay/op/strategy/x86.py
##
@@ -84,8 +88,13 @@ def conv2d_strategy_cpu(attrs, inputs, out_type, target):
raise
Menooker commented on pull request #5357:
URL: https://github.com/apache/incubator-tvm/pull/5357#issuecomment-621165175
@icemelon9
If you unchenge these, you will get
```
File "/home/menooker/work/mobilenet/tvm/python/tvm/relay/op/nn/_nn.py",
line 97, in
mbaret commented on a change in pull request #5345:
URL: https://github.com/apache/incubator-tvm/pull/5345#discussion_r417275917
##
File path: python/tvm/relay/qnn/op/legalizations.py
##
@@ -20,8 +20,8 @@
import tvm
from tvm import relay
+import numpy as np
Review
This is an automated email from the ASF dual-hosted git repository.
zhaowu pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git
The following commit(s) were added to refs/heads/master by this push:
new 063ba63 [Frontend][TFLite] support for
maheshambule opened a new pull request #5474:
URL: https://github.com/apache/incubator-tvm/pull/5474
- added support for ADD_N operator
- refactored some of the ADD code
This is an automated message from the Apache Git
siju-samuel commented on pull request #5472:
URL: https://github.com/apache/incubator-tvm/pull/5472#issuecomment-621154395
`
Running relay Tensorflow frontend test...
= test session starts
==
platform linux -- Python 3.6.10,
80 matches
Mail list logo
