benjamim93 opened a new pull request #5381:
URL: https://github.com/apache/incubator-tvm/pull/5381
set_output is enabled to redirect the output, with unnecessary copies saved.
It can be regarded as the output version of set_input_zero_copy.
u99127 commented on issue #5355:
URL: https://github.com/apache/incubator-tvm/pull/5355#issuecomment-616392562
> > @siju-samuel is correct. If we do like this pr, when we `import tvm`,
the users will be required to install tflite package, which is not we want to
see.
>
> I missed th
roastduck opened a new pull request #5382:
URL: https://github.com/apache/incubator-tvm/pull/5382
Suppose the extent of `threadIdx.x` equals to the warp size, only
`threadIdx.x` should be a index of a warp memory, while `threadIdx.y` or
`threadIdx.z` should not, since warp shuffle is insid
benjamim93 commented on issue #5381:
URL: https://github.com/apache/incubator-tvm/pull/5381#issuecomment-616424890
Review request @hlu1 @zhiics
This is an automated message from the Apache Git Service.
To respond to the mess
maheshambule commented on issue #5330:
URL: https://github.com/apache/incubator-tvm/pull/5330#issuecomment-616454075
@siju-samuel, Thanks.
@kevinthesun, Please help in review and merge.
This is an automated message from t
mbaret commented on issue #5360:
URL: https://github.com/apache/incubator-tvm/pull/5360#issuecomment-616497178
Is there somewhere that QnnCanonicalize is tested where I can add a test?
This is an automated message from the Ap
siju-samuel opened a new pull request #5383:
URL: https://github.com/apache/incubator-tvm/pull/5383
Pytorch frontend op support
- where
- addcdiv
- addcmul
@masahi please help to review this pytorch ops. Thanks.
--
leandron commented on issue #4953:
URL: https://github.com/apache/incubator-tvm/issues/4953#issuecomment-616610699
We applied a workaround, pinning the xgboost version to be 0.90. Which
XGBoost version you see in the image you created from scratch?
mbaret commented on issue #5345:
URL: https://github.com/apache/incubator-tvm/pull/5345#issuecomment-616650764
Does anyone have a chance to take a look at this?
This is an automated message from the Apache Git Service.
To res
tqchen commented on issue #5384:
URL: https://github.com/apache/incubator-tvm/issues/5384#issuecomment-616660386
cc @yzhliu
This is an automated message from the Apache Git Service.
To respond to the message, please log on t
tqchen opened a new issue #5384:
URL: https://github.com/apache/incubator-tvm/issues/5384
See
-
https://github.com/apache/incubator-tvm/blob/master/src/arith/modular_set.cc#L305
- https://github.com/apache/incubator-tvm/blob/master/src/arith/util.cc#L29
Consider move the impl
anijain2305 commented on issue #5363:
URL: https://github.com/apache/incubator-tvm/pull/5363#issuecomment-616705895
@cbalint13 @masahi PTAL again. I relaxed to have any type of floats. And
also added the Top-5 labels for few images in the desciption.
--
anijain2305 commented on issue #5360:
URL: https://github.com/apache/incubator-tvm/pull/5360#issuecomment-616720216
You can add it here - tests/python/relay/test_op_qnn_concatenate.py
It calls QNNCanonicalize internally
Th
wpan11nv edited a comment on issue #5214:
URL: https://github.com/apache/incubator-tvm/pull/5214#issuecomment-616723522
> @wpan11nv @yongfeng-nv can you suggest a bit about possible optimizations
that can be done?
The cuda schedule emits 4 kernels, which cause lots of IO overhead.
wpan11nv commented on issue #5214:
URL: https://github.com/apache/incubator-tvm/pull/5214#issuecomment-616723522
> @wpan11nv @yongfeng-nv can you suggest a bit about possible optimizations
that can be done?
The cuda schedule emit 4 kernels, which cause lots of IO overhead. Ideally,
zhiics edited a comment on issue #5345:
URL: https://github.com/apache/incubator-tvm/pull/5345#issuecomment-616725734
I am actually not quite encouraging us to introduce this to the relay
namespace as this file only really has one function and it is rarely used
---
zhiics commented on issue #5345:
URL: https://github.com/apache/incubator-tvm/pull/5345#issuecomment-616725734
I am actually not quite encourage us to introduce this to the relay
namespace as this file only really has one function and it is rarely used
Somisary removed a comment on issue #5345:
URL: https://github.com/apache/incubator-tvm/pull/5345#issuecomment-616730802
Where would be appropriate? An alternative would be that given this is used
only in two places we could inline it into those files.
Somisary commented on issue #5345:
URL: https://github.com/apache/incubator-tvm/pull/5345#issuecomment-616730802
Where would be appropriate? An alternative would be that given this is used
only in two places we could inline it into those files.
anijain2305 commented on issue #5345:
URL: https://github.com/apache/incubator-tvm/pull/5345#issuecomment-616753221
Maybe we want to move it some other already existing file. But, I think we
need this in Relay namespace. This is used in TFLite parser and QNN Op
legalization.
~~~
zhiics commented on issue #5345:
URL: https://github.com/apache/incubator-tvm/pull/5345#issuecomment-616769231
Yeah, I meant we probably don't want to put under relay root folder,
somewhere else is okay to me.
This is an aut
zhiics edited a comment on issue #5345:
URL: https://github.com/apache/incubator-tvm/pull/5345#issuecomment-616769231
Yeah, I meant we probably don't want to put under relay root folder,
somewhere else (even some duplication in each file) is okay to me.
---
hcho3 commented on issue #4953:
URL: https://github.com/apache/incubator-tvm/issues/4953#issuecomment-616770561
@leandron I checked out commit 8502691b5b7ca152da9eb626529070db53d479c8 so
that XGBoost 1.0.x is used.
This is a
tqchen opened a new pull request #5385:
URL: https://github.com/apache/incubator-tvm/pull/5385
The legacy Simplify/CanonicalSimplify are now a thin wrapper around the
Analyzer.
This PR removes these functions and migrated every place that requires
simplification to enforce Analyzer cr
tqchen commented on issue #4617:
URL: https://github.com/apache/incubator-tvm/issues/4617#issuecomment-616779767
Move to #4647
This is an automated message from the Apache Git Service.
To respond to the message, please log on
tqchen commented on issue #5385:
URL: https://github.com/apache/incubator-tvm/pull/5385#issuecomment-616815559
cc @yzhliu @merrymercy @Hzfengsy @spectrometerHBH @ZihengJiang
This is an automated message from the Apache Git S
tqchen opened a new pull request #5386:
URL: https://github.com/apache/incubator-tvm/pull/5386
Rationale: inline is a transformation used in te to
rewrite its internal expressions. It is not a formal IRModule->IRModule
transform pass.
Also removed the python test as the test is co
yongfeng-nv commented on a change in pull request #5367:
URL: https://github.com/apache/incubator-tvm/pull/5367#discussion_r411785613
##
File path: src/arith/int_set.cc
##
@@ -372,7 +387,27 @@ class IntervalSetEvaluator :
}
IntervalSet Eval(const PrimExpr& val) {
-r
yongfeng-nv commented on issue #5367:
URL: https://github.com/apache/incubator-tvm/pull/5367#issuecomment-616885908
> It would be great if we can dissect it down to specific cases in mind and
discuss how to solve these cases, I see two kinds of problems:
>
> * (1) EvalSet need to be
tqchen opened a new pull request #5387:
URL: https://github.com/apache/incubator-tvm/pull/5387
cc @yzhliu @ZihengJiang @Hzfengsy
This is an automated message from the Apache Git Service.
To respond to the message, please log
samwyi opened a new issue #5388:
URL: https://github.com/apache/incubator-tvm/issues/5388
**Problem**: When the Vulkan runtime uses deferred kernels to do a GPU ->
GPU copy, the source vkBuffer is released before the copy command is put into
command buffer. That can cause seg fault in vulk
tqchen commented on a change in pull request #5367:
URL: https://github.com/apache/incubator-tvm/pull/5367#discussion_r411796012
##
File path: src/arith/int_set.cc
##
@@ -372,7 +387,27 @@ class IntervalSetEvaluator :
}
IntervalSet Eval(const PrimExpr& val) {
-return
tqchen commented on issue #5388:
URL: https://github.com/apache/incubator-tvm/issues/5388#issuecomment-616896622
Thanks @samwyi for reporting the problem. Perhaps an easier fix now would be
call into Synchronize directly in when FreeDataSpace is called. This is also
the default behavior in
tqchen commented on issue #5388:
URL: https://github.com/apache/incubator-tvm/issues/5388#issuecomment-616896778
cc @ajtulloch @antinucleon
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.
tqchen pushed a change to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-tvm.git.
from b8efe27 [REFACTOR][TE] Inline -> te/schedule/operation_inline.h
(#5386)
add d9cecdf [ARITH] Remove the
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 3f03869 [Blocksparse] Pipeline for lowering dense model to
sparse-dense (#5377)
add b8efe27 [REFACTOR]
jwfromm opened a new pull request #5389:
URL: https://github.com/apache/incubator-tvm/pull/5389
In cases where an operator has multiple inputs, but those inputs are
identical, `onnx_input` would add only one copy to its internal list of inputs.
This can cause errors in some fringe cases. T
jwfromm commented on issue #5389:
URL: https://github.com/apache/incubator-tvm/pull/5389#issuecomment-616919201
@masahi, @soiferj can you take a look at this PR?
This is an automated message from the Apache Git Service.
To re
tqchen commented on issue #5382:
URL: https://github.com/apache/incubator-tvm/pull/5382#issuecomment-616926958
cc @yongfeng-nv @wpan11nv @vinx13
This is an automated message from the Apache Git Service.
To respond to the mes
tqchen commented on issue #5381:
URL: https://github.com/apache/incubator-tvm/pull/5381#issuecomment-616927783
I am not sure if this version is necessary since it involves a lot of
changes to the graph runtime itself. Given that zero copy in the output can be
relatively rare, perhaps we wa
tqchen edited a comment on issue #5381:
URL: https://github.com/apache/incubator-tvm/pull/5381#issuecomment-616927783
Thanks @benjamim93 I am not sure if this version is necessary since it
involves a lot of changes to the graph runtime itself. The implication of the
output can also be conf
tqchen edited a comment on issue #5381:
URL: https://github.com/apache/incubator-tvm/pull/5381#issuecomment-616927783
I am not sure if this version is necessary since it involves a lot of
changes to the graph runtime itself. The implication of the output can also be
confusing(e.g. who own
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 4cebb1c [ARITH] Remove legacy const patt
areusch opened a new pull request #5390:
URL: https://github.com/apache/incubator-tvm/pull/5390
Fixed the test error causing #5189. None should be VoidType, also known as
TupleType([]).
This is an automated message from the
yongfeng-nv edited a comment on issue #5367:
URL: https://github.com/apache/incubator-tvm/pull/5367#issuecomment-616035227
The current behavior of IntervalSet floormod(a, b) is rough -- it returns
[0, b-1], [-b+1, b-1], or everything. It causes extra iterations to my
schedule. This is a
yongfeng-nv commented on issue #5367:
URL: https://github.com/apache/incubator-tvm/pull/5367#issuecomment-616964958
> In particular, my comment wrt to EvalSet means we should directly use the
IntSetAnalyzer member of the analyzer, that can should have part of the Var's
bound info
Re
yongfeng-nv commented on a change in pull request #5367:
URL: https://github.com/apache/incubator-tvm/pull/5367#discussion_r411885530
##
File path: src/arith/int_set.cc
##
@@ -730,12 +765,30 @@ Map ConvertDomMap(
return dmap;
}
-IntSet EvalSet(PrimExpr e,
-
yongfeng-nv commented on a change in pull request #5367:
URL: https://github.com/apache/incubator-tvm/pull/5367#discussion_r411886246
##
File path: include/tvm/te/operation.h
##
@@ -112,6 +113,7 @@ class OperationNode : public tir::FunctionBaseNode {
const Operation& sel
u99127 opened a new pull request #5391:
URL: https://github.com/apache/incubator-tvm/pull/5391
There are a couple of testisms that are found when running the testsuite
under TF2.1. I have a separate pull request for that coming shortly.
@kevinthesun , @siju-samuel , could you please
u99127 opened a new pull request #5392:
URL: https://github.com/apache/incubator-tvm/pull/5392
This shows the changes needed to CI and the testsuite to get CI to support
TF/TFLite 1.15.2 and then on to TF / TFlite 2.1.0
There were a couple of related issues found with Tensorflow
zchuang11 commented on issue #4845:
URL: https://github.com/apache/incubator-tvm/issues/4845#issuecomment-616974448
> Due to recent situation and the current progress, we might expect a bit of
delay in the release to June/July -- we expect the unified IR refactor to land
by then. We will d
zchuang11 opened a new issue #5393:
URL: https://github.com/apache/incubator-tvm/issues/5393
Thanks for participating in the TVM community! We use https://discuss.tvm.ai
for any general usage questions and discussions. The issue tracker is used for
actionable items such as feature proposal
52 matches
Mail list logo
