junrushao commented on code in PR #12978:
URL: https://github.com/apache/tvm/pull/12978#discussion_r988577973
##
python/tvm/script/_parser/core/doc.py:
##
@@ -0,0 +1,361 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.
areusch commented on code in PR #12978:
URL: https://github.com/apache/tvm/pull/12978#discussion_r988536870
##
python/tvm/script/_parser/core/doc.py:
##
@@ -0,0 +1,361 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.
AndrewZhaoLuo commented on code in PR #12969:
URL: https://github.com/apache/tvm/pull/12969#discussion_r988521682
##
python/tvm/topi/arm_cpu/mprofile/dsp/tensordot_conv2ds.py:
##
@@ -0,0 +1,276 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more
junrushao commented on PR #12895:
URL: https://github.com/apache/tvm/pull/12895#issuecomment-1269261207
Hey @masahi, I added `executor` parameter to `extract_tasks` and
`compile_relay`, which controls the default value of
`relay.FuseOps.link_params` in pass configuration. It's quite
This is an automated email from the ASF dual-hosted git repository.
andrewzhaoluo pushed a change to branch v0.10.0
in repository https://gitbox.apache.org/repos/asf/tvm.git
from 224c347144 jostle ci
add d689733d5a update version numbers
add 485cb8804d remove vta references
This is an automated email from the ASF dual-hosted git repository.
andrewzhaoluo pushed a commit to branch v0.10.0-cleanup
in repository https://gitbox.apache.org/repos/asf/tvm.git
commit bda06171a0b6d5d73d5d1bd535427b0d04de2dd1
Author: Andrew Luo
AuthorDate: Wed Oct 5 20:00:50 2022 -0700
This is an automated email from the ASF dual-hosted git repository.
andrewzhaoluo pushed a change to branch v0.10.0-cleanup
in repository https://gitbox.apache.org/repos/asf/tvm.git
at bda06171a0 update jenksinfile
This branch includes the following new commits:
new d689733d5a
This is an automated email from the ASF dual-hosted git repository.
andrewzhaoluo pushed a commit to branch v0.10.0-cleanup
in repository https://gitbox.apache.org/repos/asf/tvm.git
commit 485cb8804d1a9480716530cf65a092a8c9424a15
Author: Andrew Luo
AuthorDate: Wed Oct 5 19:57:40 2022 -0700
This is an automated email from the ASF dual-hosted git repository.
andrewzhaoluo pushed a commit to branch v0.10.0-cleanup
in repository https://gitbox.apache.org/repos/asf/tvm.git
commit d689733d5ae3d4d16455488cbfd9092c6f065029
Author: Andrew Luo
AuthorDate: Wed Oct 5 10:50:58 2022 -0700
This is an automated email from the ASF dual-hosted git repository.
junrushao pushed a change to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git
from 59b945742d [FIX,AUTOSCHEDULER,METASCHEDULE] Handle negative extents in
featurization (#12990)
add 3d5e440fff [ci]
junrushao merged PR #12956:
URL: https://github.com/apache/tvm/pull/12956
--
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 specific comment.
To unsubscribe, e-mail:
masahi commented on issue #12987:
URL: https://github.com/apache/tvm/issues/12987#issuecomment-1269137655
It compiles fine on my environment using clang 14. Maybe the older Apple
clang doesn't have complete support for `std::optional`. If you google "error:
'value' is unavailable:
masahi closed issue #12987: Build TVM on MacOS failed
URL: https://github.com/apache/tvm/issues/12987
--
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 specific comment.
To unsubscribe, e-mail:
zxybazh opened a new issue, #12997:
URL: https://github.com/apache/tvm/issues/12997
I noticed a data type issue that can be triggered when applying compute-at,
causing invalid expression in TIR script with incompatible datatype, thus
failure in post processors and build.
To
junrushao commented on PR #12806:
URL: https://github.com/apache/tvm/pull/12806#issuecomment-1269131468
If there is no impact to overall performance at all, as we have some initial
data, I'm happy to merge it in mainly because you prefer, i.e. more textualism
in arithmetic intensity
shingjan opened a new issue, #12996:
URL: https://github.com/apache/tvm/issues/12996
The current codegen of relay to TIR will introduce undefined variables for
topi implementation lie mentioned in #12993. This
This is an automated email from the ASF dual-hosted git repository.
junrushao pushed a change to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git
from d023ef47fc [TIR, MetaSchedule] Preserve unit block iters for
auto-tensorization (#12974)
add 59b945742d
tvm-bot commented on PR #12994:
URL: https://github.com/apache/tvm/pull/12994#issuecomment-1269126732
Thanks for contributing to TVM! Please refer to the contributing guidelines
https://tvm.apache.org/docs/contribute/ for useful information and tips. Please
request code reviews
junrushao merged PR #12990:
URL: https://github.com/apache/tvm/pull/12990
--
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 specific comment.
To unsubscribe, e-mail:
This is an automated email from the ASF dual-hosted git repository.
github-bot pushed a change to branch nightly-docker-update
in repository https://gitbox.apache.org/repos/asf/tvm.git
discard 438fdd2477 [ci][docker] Nightly Docker image update
add a997c23e94 [CODEGEN][OPENCL] Sampler
shingjan opened a new issue, #12995:
URL: https://github.com/apache/tvm/issues/12995
The following TIR generated from the `executor_codegen_` of relay is not
valid and therefore shouldn't be accepted as a valid input to TVM's relay
executor.
```
@tvm.script.ir_module
class
junrushao commented on PR #12990:
URL: https://github.com/apache/tvm/pull/12990#issuecomment-1269125842
would be great if we could fix the source of the problem, as you indicated,
the lower bound inferred. we do assume the IR is valid in the feature extractor
--
This is an automated
junrushao commented on code in PR #12978:
URL: https://github.com/apache/tvm/pull/12978#discussion_r988443323
##
python/tvm/script/_parser/core/doc.py:
##
@@ -0,0 +1,361 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.
junrushao commented on code in PR #12978:
URL: https://github.com/apache/tvm/pull/12978#discussion_r988443323
##
python/tvm/script/_parser/core/doc.py:
##
@@ -0,0 +1,361 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.
tvm-bot commented on PR #12994:
URL: https://github.com/apache/tvm/pull/12994#issuecomment-1269121591
Thanks for contributing to TVM! Please refer to the contributing guidelines
https://tvm.apache.org/docs/contribute/ for useful information and tips. Please
request code reviews
shingjan commented on PR #12027:
URL: https://github.com/apache/tvm/pull/12027#issuecomment-1269116473
closed with #12993
--
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 specific comment.
shingjan closed pull request #11885: Torch embedding bag
URL: https://github.com/apache/tvm/pull/11885
--
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 specific comment.
To unsubscribe, e-mail:
shingjan commented on PR #11885:
URL: https://github.com/apache/tvm/pull/11885#issuecomment-1269116066
closed with @12993
--
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 specific comment.
shingjan closed pull request #12027: [PyTorch] Add aten::embedding_bag
URL: https://github.com/apache/tvm/pull/12027
--
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 specific comment.
To
shingjan opened a new pull request, #12993:
URL: https://github.com/apache/tvm/pull/12993
This PR is in draft as it is blocked by:
This PR intends to add `embedding_bag` for relay/topi and is related to
@12027 and #12485.
Co-authored by: Masahiro Masuda
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r988422820
##
python/tvm/meta_schedule/relay_integration.py:
##
@@ -69,47 +124,229 @@ def extract_task_from_relay(
"""
# pylint: disable=import-outside-toplevel
This is an automated email from the ASF dual-hosted git repository.
github-bot pushed a change to branch last-successful
in repository https://gitbox.apache.org/repos/asf/tvm.git
from 2e257f0376 [Hexagon] Do not use `target` test fixture in Hexagon tests
(#12981)
add 2860a50ff9
This is an automated email from the ASF dual-hosted git repository.
github-bot pushed a change to branch last-successful
in repository https://gitbox.apache.org/repos/asf/tvm.git
from 9618e6a457 [TIR] Use IndexMap to transform NDArray (#12949)
add 393d5f7fd2 [FIX,PROFILER] Fix PAPI
areusch commented on code in PR #12978:
URL: https://github.com/apache/tvm/pull/12978#discussion_r988412765
##
python/tvm/script/_parser/core/doc.py:
##
@@ -0,0 +1,361 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.
tkonolige commented on PR #12990:
URL: https://github.com/apache/tvm/pull/12990#issuecomment-1269048060
Yes, there is an example where the inferred lower bound is less than zero. I
cannot share the model though.
--
This is an automated message from the Apache Git Service.
To respond to
This is an automated email from the ASF dual-hosted git repository.
junrushao pushed a change to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git
from 2860a50ff9 [Hexagon] Fix RPC session close by adding shutdown
PackedFunc (#12960)
add d023ef47fc [TIR,
junrushao merged PR #12974:
URL: https://github.com/apache/tvm/pull/12974
--
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 specific comment.
To unsubscribe, e-mail:
junrushao commented on PR #12990:
URL: https://github.com/apache/tvm/pull/12990#issuecomment-1269035712
Well…it’s definitely true that both systems didn’t consider negative
extents is there any concrete case where loops are of negative extents?
--
This is an automated message from the
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r988367063
##
python/tvm/meta_schedule/schedule_rule/schedule_rule.py:
##
@@ -76,6 +81,35 @@ def clone(self) -> "ScheduleRule":
"""
return
csullivan commented on code in PR #12863:
URL: https://github.com/apache/tvm/pull/12863#discussion_r988193162
##
include/tvm/arith/analyzer.h:
##
@@ -275,6 +275,36 @@ class RewriteSimplifier {
*/
std::function EnterConstraint(const PrimExpr& constraint);
+ /*! \brief
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r988351862
##
src/meta_schedule/mutator/mutator.cc:
##
@@ -72,6 +97,10 @@ TVM_REGISTER_GLOBAL("meta_schedule.MutatorApply")
});
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r988351447
##
src/meta_schedule/postproc/postproc.cc:
##
@@ -67,6 +109,10 @@
TVM_REGISTER_GLOBAL("meta_schedule.PostprocInitializeWithTuneContext")
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r988351184
##
python/tvm/meta_schedule/postproc/postproc.py:
##
@@ -70,6 +74,35 @@ def clone(self) -> "Postproc":
"""
return _ffi_api.PostprocClone(self) #
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r988350143
##
src/meta_schedule/postproc/postproc.cc:
##
@@ -50,6 +50,48 @@ Postproc Postproc::PyPostproc(
return Postproc(n);
}
+Array Postproc::DefaultLLVM() {
+ return
jverma-quic commented on code in PR #12971:
URL: https://github.com/apache/tvm/pull/12971#discussion_r988348030
##
src/driver/driver_api.cc:
##
@@ -252,6 +255,10 @@ Array CreatePassList(bool
disable_loop_partition) {
pass_list.push_back(
jverma-quic commented on code in PR #12971:
URL: https://github.com/apache/tvm/pull/12971#discussion_r988347062
##
src/runtime/hexagon/profiler/README.md:
##
@@ -0,0 +1,152 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+# Hexagon lightweight instrumentation based profiling (LWP)
+
+For
jverma-quic commented on code in PR #12971:
URL: https://github.com/apache/tvm/pull/12971#discussion_r988345984
##
src/runtime/hexagon/profiler/README.md:
##
@@ -0,0 +1,152 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+# Hexagon lightweight instrumentation based profiling (LWP)
+
+For
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r988344383
##
python/tvm/meta_schedule/database/database.py:
##
@@ -361,6 +363,44 @@ def current() -> Optional["Database"]:
"""Get the current database under scope."""
jverma-quic commented on code in PR #12971:
URL: https://github.com/apache/tvm/pull/12971#discussion_r988341928
##
src/runtime/hexagon/profiler/lwp_handler.S:
##
@@ -0,0 +1,87 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
jverma-quic commented on code in PR #12971:
URL: https://github.com/apache/tvm/pull/12971#discussion_r988341680
##
src/tir/transforms/profile_instrumentation.cc:
##
@@ -0,0 +1,287 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor
janetsc opened a new pull request, #12992:
URL: https://github.com/apache/tvm/pull/12992
Update HexagonVtcmPool to use compute_resource_query_VTCM to query for the
total and available blocks in VTCM.
--
This is an automated message from the Apache Git Service.
To respond to the message,
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r988341530
##
python/tvm/meta_schedule/cost_model/cost_model.py:
##
@@ -97,6 +104,30 @@ def predict(self, context: TuneContext, candidates:
List[MeasureCandidate]) -> n
mehrdadh merged PR #12960:
URL: https://github.com/apache/tvm/pull/12960
--
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 specific comment.
To unsubscribe, e-mail:
This is an automated email from the ASF dual-hosted git repository.
mehrdadh pushed a change to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git
from 2e257f0376 [Hexagon] Do not use `target` test fixture in Hexagon tests
(#12981)
add 2860a50ff9 [Hexagon] Fix RPC
guberti commented on code in PR #12969:
URL: https://github.com/apache/tvm/pull/12969#discussion_r988338105
##
python/tvm/topi/arm_cpu/mprofile/dsp/micro_kernel/tensordot.py:
##
@@ -0,0 +1,136 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more
jverma-quic commented on PR #12971:
URL: https://github.com/apache/tvm/pull/12971#issuecomment-1268953886
> @jverma-quic this is pretty awesome! I've wanted something like this for a
while.
>
> Thinking longer-term, I'd like to see this supported for more targets than
hexagon. Maybe
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r988334549
##
python/tvm/meta_schedule/builder/builder.py:
##
@@ -126,6 +128,22 @@ def build(self, build_inputs: List[BuilderInput]) ->
List[BuilderResult]:
"""
guberti commented on code in PR #12969:
URL: https://github.com/apache/tvm/pull/12969#discussion_r988333507
##
tests/python/relay/strategy/arm_cpu/test_conv2d_nhwc.py:
##
@@ -150,5 +147,38 @@ class TestConv2d_HWIO(BasicConv2dTests):
schedule_name =
masahi opened a new pull request, #12991:
URL: https://github.com/apache/tvm/pull/12991
This lets us enable `RewriteLayout` for Hexagon.
--
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
guberti commented on code in PR #12969:
URL: https://github.com/apache/tvm/pull/12969#discussion_r988325948
##
python/tvm/topi/arm_cpu/mprofile/dsp/tensordot_conv2ds.py:
##
@@ -0,0 +1,276 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor
guberti commented on code in PR #12969:
URL: https://github.com/apache/tvm/pull/12969#discussion_r988325948
##
python/tvm/topi/arm_cpu/mprofile/dsp/tensordot_conv2ds.py:
##
@@ -0,0 +1,276 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r988321661
##
python/tvm/contrib/torch/optimize_torch.py:
##
@@ -62,75 +51,32 @@ def forward(self, *torch_inputs: Tuple[torch.Tensor]):
return ret
-def llvm_target():
tkonolige opened a new pull request, #12990:
URL: https://github.com/apache/tvm/pull/12990
Both auto_scheduler and metaschedule featurization did not lower bound loop
extents by zero, leading to negative bounds. When multiplied with element size
integer underflow occurred before converting
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r98832
##
python/tvm/contrib/torch/optimize_torch.py:
##
@@ -62,75 +51,32 @@ def forward(self, *torch_inputs: Tuple[torch.Tensor]):
return ret
-def llvm_target():
mehrdadh merged PR #12981:
URL: https://github.com/apache/tvm/pull/12981
--
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 specific comment.
To unsubscribe, e-mail:
This is an automated email from the ASF dual-hosted git repository.
mehrdadh pushed a change to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git
from 393d5f7fd2 [FIX,PROFILER] Fix PAPI test and test in CI (#12953)
add 2e257f0376 [Hexagon] Do not use `target` test
guberti commented on code in PR #12969:
URL: https://github.com/apache/tvm/pull/12969#discussion_r988317552
##
python/tvm/topi/arm_cpu/mprofile/dsp/micro_kernel/tensordot.py:
##
@@ -0,0 +1,136 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more
masahi commented on code in PR #12980:
URL: https://github.com/apache/tvm/pull/12980#discussion_r988310496
##
python/tvm/testing/utils.py:
##
@@ -981,6 +982,50 @@ def _corstone300_compile_time_check():
requires_vitis_ai = Feature("vitis_ai", "Vitis AI",
masahi commented on code in PR #12980:
URL: https://github.com/apache/tvm/pull/12980#discussion_r988310496
##
python/tvm/testing/utils.py:
##
@@ -981,6 +982,50 @@ def _corstone300_compile_time_check():
requires_vitis_ai = Feature("vitis_ai", "Vitis AI",
janetsc commented on code in PR #12947:
URL: https://github.com/apache/tvm/pull/12947#discussion_r986051570
##
src/runtime/hexagon/hexagon_vtcm_pool.h:
##
@@ -0,0 +1,100 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
adstraw commented on code in PR #12954:
URL: https://github.com/apache/tvm/pull/12954#discussion_r986077448
##
tests/python/contrib/test_hexagon/test_software_pipeline_async.py:
##
@@ -46,18 +47,33 @@ def plus_one_ref(a):
return plus_one_primfunc, plus_one_ref
lhutton1 commented on code in PR #12970:
URL: https://github.com/apache/tvm/pull/12970#discussion_r986067418
##
src/relay/backend/contrib/ethosn/ethosn_api.cc:
##
@@ -213,27 +227,29 @@ EthosnError EthosnAPI::QnnFullyConnected(const Expr&
expr, FullyConnectedParams*
tmoreau89 commented on PR #12971:
URL: https://github.com/apache/tvm/pull/12971#issuecomment-1265833312
CC @tkonolige
--
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 specific comment.
To
adstraw commented on code in PR #12954:
URL: https://github.com/apache/tvm/pull/12954#discussion_r986049998
##
tests/python/contrib/test_hexagon/test_software_pipeline_async.py:
##
@@ -46,18 +47,33 @@ def plus_one_ref(a):
return plus_one_primfunc, plus_one_ref
janetsc commented on code in PR #12947:
URL: https://github.com/apache/tvm/pull/12947#discussion_r985777191
##
src/runtime/hexagon/hexagon_vtcm_pool.cc:
##
@@ -0,0 +1,166 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
jverma-quic opened a new pull request, #12971:
URL: https://github.com/apache/tvm/pull/12971
This's done by instrumenting the code with profiling builtin calls using a
TIR pass. During codegen, these builtin calls are replaced with the calls to a
hexagon specific handler which records the
mehrdadh commented on PR #12818:
URL: https://github.com/apache/tvm/pull/12818#issuecomment-1265804656
@areusch I think this is ready, PTAL. thanks!
--
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
kparzysz-quic closed pull request #12957: [Hexagon] Remove TVM logging from
performance-sensitive parts of runtime
URL: https://github.com/apache/tvm/pull/12957
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL
kparzysz-quic commented on PR #12957:
URL: https://github.com/apache/tvm/pull/12957#issuecomment-1265790665
Closing this, will come up with a smaller fix.
--
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
tmoreau89 commented on code in PR #12954:
URL: https://github.com/apache/tvm/pull/12954#discussion_r986037292
##
tests/python/contrib/test_hexagon/test_software_pipeline_async.py:
##
@@ -46,18 +47,33 @@ def plus_one_ref(a):
return plus_one_primfunc, plus_one_ref
ashutosh-arm commented on code in PR #12970:
URL: https://github.com/apache/tvm/pull/12970#discussion_r985985300
##
tests/python/contrib/test_ethosn/test_fullyconnected.py:
##
@@ -30,7 +32,11 @@ def _get_model(
):
"""Return a model an any parameters it may have"""
a
sunggg commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r985956777
##
python/tvm/meta_schedule/database/database.py:
##
@@ -361,6 +363,44 @@ def current() -> Optional["Database"]:
"""Get the current database under scope."""
tkonolige commented on code in PR #12959:
URL: https://github.com/apache/tvm/pull/12959#discussion_r985993948
##
tests/scripts/task_config_build_cpu.sh:
##
@@ -55,6 +55,7 @@ echo set\(USE_CMSISNN OFF\) >> config.cmake
echo set\(USE_VITIS_AI ON\) >> config.cmake
echo
janetsc commented on code in PR #12947:
URL: https://github.com/apache/tvm/pull/12947#discussion_r985990771
##
src/runtime/hexagon/hexagon_vtcm_pool.cc:
##
@@ -0,0 +1,166 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
ashutosh-arm commented on code in PR #12970:
URL: https://github.com/apache/tvm/pull/12970#discussion_r985981792
##
src/relay/backend/contrib/ethosn/ethosn_api.cc:
##
@@ -213,27 +227,29 @@ EthosnError EthosnAPI::QnnFullyConnected(const Expr&
expr, FullyConnectedParams*
Lunderberg commented on PR #12942:
URL: https://github.com/apache/tvm/pull/12942#issuecomment-1265691413
That makes sense, and agreed as a path forward. Closing this PR until and
unless it becomes necessary or the performance difference becomes much larger.
--
This is an automated
Lunderberg closed pull request #12942: [Arith][Refactor] Extract And/Or/Not
handling from RewriteSimplifier
URL: https://github.com/apache/tvm/pull/12942
--
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
janetsc commented on PR #12957:
URL: https://github.com/apache/tvm/pull/12957#issuecomment-1265677915
>
Ah, I see some tests are throwing this even for runtime_hexbuffs (before
ReleaseResources is called.). I'm going to continue to investigate.
--
This is an automated message
Lunderberg commented on PR #12949:
URL: https://github.com/apache/tvm/pull/12949#issuecomment-1265673499
Good points. I'd been thinking in terms of a small number of discontiguous
breaks between largely contiguous regions, but that probably would be rather
rare. Agreed that it isn't
kparzysz-quic commented on PR #12957:
URL: https://github.com/apache/tvm/pull/12957#issuecomment-1265647363
> I tested this locally, which I prefer to removing the assert in
`FreeHexagonBuffer`. This will catch the error if we are operating on the
static buffer manager:
This still
kparzysz-quic commented on code in PR #12947:
URL: https://github.com/apache/tvm/pull/12947#discussion_r985911415
##
src/runtime/hexagon/hexagon_vtcm_pool.cc:
##
@@ -0,0 +1,166 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
janetsc commented on code in PR #12947:
URL: https://github.com/apache/tvm/pull/12947#discussion_r985924362
##
src/runtime/hexagon/hexagon_vtcm_pool.h:
##
@@ -0,0 +1,100 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
junrushao commented on code in PR #12895:
URL: https://github.com/apache/tvm/pull/12895#discussion_r985899276
##
python/tvm/meta_schedule/relay_integration.py:
##
@@ -15,28 +15,82 @@
# specific language governing permissions and limitations
# under the License.
lhutton1 opened a new pull request, #12970:
URL: https://github.com/apache/tvm/pull/12970
The NPU driver stack expects weights in IO (HWIO) format, however, Relay
uses an OI representation. Although the shape of the weight tensor was
correctly changed during codegen, the values in the
hpanda-naut commented on code in PR #12898:
URL: https://github.com/apache/tvm/pull/12898#discussion_r985883193
##
.github/ISSUE_TEMPLATE/bug-report.md:
##
@@ -1,8 +1,8 @@
---
name: "\U0001F41B Bug report"
-about: To help the developer act on the issues, please include a
janetsc commented on PR #12957:
URL: https://github.com/apache/tvm/pull/12957#issuecomment-1265594905
> The degradation started a while back, but we didn't notice, because the
tests were passing without timing out. The main contributor to the degradation
is the exception being thrown, but
kparzysz-quic commented on PR #12957:
URL: https://github.com/apache/tvm/pull/12957#issuecomment-1265559044
The degradation started a while back, but we didn't notice, because the
tests were passing without timing out. The main contributor to the degradation
is the exception being thrown,
Lunderberg commented on code in PR #12947:
URL: https://github.com/apache/tvm/pull/12947#discussion_r985852507
##
src/runtime/hexagon/hexagon_vtcm_pool.h:
##
@@ -0,0 +1,100 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
janetsc commented on PR #12957:
URL: https://github.com/apache/tvm/pull/12957#issuecomment-1265527449
>
> Our local simulator tests would time out after 1h, now they finish within
30 min. I used a subset of the fp16 conv tests, and without this change they
would take about
janetsc commented on code in PR #12947:
URL: https://github.com/apache/tvm/pull/12947#discussion_r985768688
##
src/runtime/hexagon/hexagon_vtcm_pool.h:
##
@@ -0,0 +1,100 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
1 - 100 of 127 matches
Mail list logo
