[GitHub] [tvm] leowang1225 commented on a change in pull request #7185: [AutoScheduler] Add custom build function
leowang1225 commented on a change in pull request #7185:
URL: https://github.com/apache/tvm/pull/7185#discussion_r551162129
##
File path: python/tvm/auto_scheduler/measure.py
##
@@ -624,12 +647,7 @@ def local_build_worker(args):
The build result of this Builder thread.
"""
inp, build_func, timeout, verbose = args
-if build_func == "default":
-build_func = tar.tar
-elif build_func == "ndk":
-build_func = ndk.create_shared
-else:
-raise ValueError("Invalid build_func" + build_func)
+build_func = BuildFunc.build_func
Review comment:
custom build func is a python callable object, and it has attribute and
so on, we can not serialize it into args.
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
[GitHub] [tvm] cxcxcxcx commented on a change in pull request #7190: Makes sure g_last_error is null terminated.
cxcxcxcx commented on a change in pull request #7190:
URL: https://github.com/apache/tvm/pull/7190#discussion_r551159043
##
File path: src/runtime/crt/common/crt_runtime_api.c
##
@@ -38,7 +38,10 @@
static char g_last_error[1024];
-void TVMAPISetLastError(const char* msg) { strncpy(g_last_error, msg,
sizeof(g_last_error)); }
+void TVMAPISetLastError(const char* msg) {
+ strncpy(g_last_error, msg, sizeof(g_last_error) - 1);
+ g_last_error[sizeof(g_last_error) - 1] = 0;
Review comment:
Does "truncate msg to the size of g_last_error" mean mutating "msg"?
That doesn't sound always doable. My understanding:
1. Before the proposed change, when the error message is longer than 1024,
g_last_error won't be null terminated. I believe this is what the warning is
about.
2. With the change, g_last_error will always be null-terminated.
3. If we use `strcpy`, we need to strlen first. If the length is long, we
need to do something special. I don't think it's trivial. It could be faster,
but I wonder how much it matters, since errors are probably rare.
4. snprintf may be an alternative. But probably not too different in the
context.
Given these, would you reconsider the change as is?
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
[GitHub] [tvm] liangfu commented on a change in pull request #7190: Makes sure g_last_error is null terminated.
liangfu commented on a change in pull request #7190:
URL: https://github.com/apache/tvm/pull/7190#discussion_r551154025
##
File path: src/runtime/crt/common/crt_runtime_api.c
##
@@ -38,7 +38,10 @@
static char g_last_error[1024];
-void TVMAPISetLastError(const char* msg) { strncpy(g_last_error, msg,
sizeof(g_last_error)); }
+void TVMAPISetLastError(const char* msg) {
+ strncpy(g_last_error, msg, sizeof(g_last_error) - 1);
+ g_last_error[sizeof(g_last_error) - 1] = 0;
Review comment:
I think no null-character is implicitly appended at the end of
destination only if source is longer than num, otherwise, destination is padded
with zeros until a total of num characters have been written to it. See
[strncpy - C++ reference](http://www.cplusplus.com/reference/cstring/strncpy/).
The error reported in the compiler meant "bound 1024 equals destination
size". As a correction for this, I think it'll be better to copy msg directly
to g_last_error:
```c
strcpy(g_last_error, msg);
```
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
[GitHub] [tvm] FrozenGene commented on pull request #7185: [AutoScheduler] Add custom build function
FrozenGene commented on pull request #7185: URL: https://github.com/apache/tvm/pull/7185#issuecomment-753811336 @jcf94 could you have another one round of review? 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] leowang1225 commented on a change in pull request #7185: [AutoScheduler] Add custom build function
leowang1225 commented on a change in pull request #7185: URL: https://github.com/apache/tvm/pull/7185#discussion_r551153793 ## File path: python/tvm/auto_scheduler/measure.py ## @@ -652,7 +652,6 @@ def local_build_worker(args): The build result of this Builder thread. """ inp, build_func, timeout, verbose = args -assert any(build_func == name for name in BuildFunc.name) Review comment: any is error, so I change it 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] FrozenGene commented on a change in pull request #7185: [AutoScheduler] Add custom build function
FrozenGene commented on a change in pull request #7185: URL: https://github.com/apache/tvm/pull/7185#discussion_r551149875 ## File path: python/tvm/auto_scheduler/measure.py ## @@ -652,7 +652,6 @@ def local_build_worker(args): The build result of this Builder thread. """ inp, build_func, timeout, verbose = args -assert any(build_func == name for name in BuildFunc.name) Review comment: it is removed by `black` tool automatically? 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] FrozenGene commented on a change in pull request #7185: [AutoScheduler] Add custom build function
FrozenGene commented on a change in pull request #7185: URL: https://github.com/apache/tvm/pull/7185#discussion_r551141295 ## File path: python/tvm/auto_scheduler/measure.py ## @@ -303,12 +312,26 @@ class LocalBuilder(ProgramBuilder): This is used in a wrapper of the multiprocessing.Process.join(). n_parallel : int = multiprocessing.cpu_count() Number of threads used to build in parallel. -build_func : str = 'default' -The name of registered build function. +build_func: callable or str = "default" +If is 'default', use default build function +If is 'ndk', use function for android ndk +If is callable, use it as custom build function, expect lib_format field. """ def __init__(self, timeout=15, n_parallel=multiprocessing.cpu_count(), build_func="default"): -self.__init_handle_by_constructor__(_ffi_api.LocalBuilder, timeout, n_parallel, build_func) +if build_func == "default": Review comment: He should keep the state information of `build_func` between functions, so just `self.build_func` can not meet the requirement because we will create one class instance. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] leowang1225 commented on a change in pull request #7185: [AutoScheduler] Add custom build function
leowang1225 commented on a change in pull request #7185: URL: https://github.com/apache/tvm/pull/7185#discussion_r551140443 ## File path: python/tvm/auto_scheduler/measure.py ## @@ -303,12 +312,26 @@ class LocalBuilder(ProgramBuilder): This is used in a wrapper of the multiprocessing.Process.join(). n_parallel : int = multiprocessing.cpu_count() Number of threads used to build in parallel. -build_func : str = 'default' -The name of registered build function. +build_func: callable or str = "default" +If is 'default', use default build function +If is 'ndk', use function for android ndk +If is callable, use it as custom build function, expect lib_format field. """ def __init__(self, timeout=15, n_parallel=multiprocessing.cpu_count(), build_func="default"): -self.__init_handle_by_constructor__(_ffi_api.LocalBuilder, timeout, n_parallel, build_func) +if build_func == "default": Review comment: when auto_scheduler.local_builder.build is called, the context can't get LocalBuilder class instance 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] FrozenGene commented on a change in pull request #7185: [AutoScheduler] Add custom build function
FrozenGene commented on a change in pull request #7185:
URL: https://github.com/apache/tvm/pull/7185#discussion_r551139727
##
File path: python/tvm/auto_scheduler/measure.py
##
@@ -624,12 +647,7 @@ def local_build_worker(args):
The build result of this Builder thread.
"""
inp, build_func, timeout, verbose = args
-if build_func == "default":
-build_func = tar.tar
-elif build_func == "ndk":
-build_func = ndk.create_shared
-else:
-raise ValueError("Invalid build_func" + build_func)
+build_func = BuildFunc.build_func
Review comment:
If this is a custom build func, we can not serialize it into args,
because it is a callable object, not one str.
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
[GitHub] [tvm] FrozenGene commented on a change in pull request #7185: [AutoScheduler] Add custom build function
FrozenGene commented on a change in pull request #7185:
URL: https://github.com/apache/tvm/pull/7185#discussion_r551139319
##
File path: python/tvm/auto_scheduler/measure.py
##
@@ -624,12 +647,7 @@ def local_build_worker(args):
The build result of this Builder thread.
"""
inp, build_func, timeout, verbose = args
-if build_func == "default":
-build_func = tar.tar
-elif build_func == "ndk":
-build_func = ndk.create_shared
-else:
-raise ValueError("Invalid build_func" + build_func)
Review comment:
we could add one check `assert any(build_func == name for name in
BuildFunc.nam)`
##
File path: python/tvm/auto_scheduler/measure.py
##
@@ -63,6 +63,15 @@
# We use 1e10 instead of sys.float_info.max for better readability in log
MAX_FLOAT = 1e10
+class BuildFunc:
+""" store build_func name and callable to class variable.
Review comment:
No need of `to class variable`
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
[GitHub] [tvm] jcf94 commented on a change in pull request #7185: [AutoScheduler] Add custom build function
jcf94 commented on a change in pull request #7185:
URL: https://github.com/apache/tvm/pull/7185#discussion_r551138247
##
File path: python/tvm/auto_scheduler/measure.py
##
@@ -624,12 +647,7 @@ def local_build_worker(args):
The build result of this Builder thread.
"""
inp, build_func, timeout, verbose = args
-if build_func == "default":
-build_func = tar.tar
-elif build_func == "ndk":
-build_func = ndk.create_shared
-else:
-raise ValueError("Invalid build_func" + build_func)
+build_func = BuildFunc.build_func
Review comment:
Get `build_func` from the `args`, rather than get from the global
`BuildFunc`.
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
[GitHub] [tvm] jcf94 commented on a change in pull request #7185: [AutoScheduler] Add custom build function
jcf94 commented on a change in pull request #7185: URL: https://github.com/apache/tvm/pull/7185#discussion_r551137236 ## File path: python/tvm/auto_scheduler/measure.py ## @@ -303,12 +312,26 @@ class LocalBuilder(ProgramBuilder): This is used in a wrapper of the multiprocessing.Process.join(). n_parallel : int = multiprocessing.cpu_count() Number of threads used to build in parallel. -build_func : str = 'default' -The name of registered build function. +build_func: callable or str = "default" +If is 'default', use default build function +If is 'ndk', use function for android ndk +If is callable, use it as custom build function, expect lib_format field. """ def __init__(self, timeout=15, n_parallel=multiprocessing.cpu_count(), build_func="default"): -self.__init_handle_by_constructor__(_ffi_api.LocalBuilder, timeout, n_parallel, build_func) +if build_func == "default": Review comment: So this is designed to use the class static member? Why not just use a `self.build_func`. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] leowang1225 commented on a change in pull request #7185: [AutoScheduler] Add custom build function
leowang1225 commented on a change in pull request #7185: URL: https://github.com/apache/tvm/pull/7185#discussion_r551135097 ## File path: python/tvm/auto_scheduler/measure.py ## @@ -63,6 +63,9 @@ # We use 1e10 instead of sys.float_info.max for better readability in log MAX_FLOAT = 1e10 +class CustomBuildFunc: Review comment: done 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] leowang1225 commented on a change in pull request #7185: [AutoScheduler] Add custom build function
leowang1225 commented on a change in pull request #7185: URL: https://github.com/apache/tvm/pull/7185#discussion_r551133295 ## File path: python/tvm/auto_scheduler/measure.py ## @@ -303,11 +306,16 @@ class LocalBuilder(ProgramBuilder): This is used in a wrapper of the multiprocessing.Process.join(). n_parallel : int = multiprocessing.cpu_count() Number of threads used to build in parallel. -build_func : str = 'default' -The name of registered build function. +build_func: callable or str Review comment: I have added BuildFunc default name = "default" and default func callable = tar.tar 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] leowang1225 commented on a change in pull request #7185: [AutoScheduler] Add custom build function
leowang1225 commented on a change in pull request #7185: URL: https://github.com/apache/tvm/pull/7185#discussion_r551132967 ## File path: python/tvm/auto_scheduler/measure.py ## @@ -628,6 +636,8 @@ def local_build_worker(args): build_func = tar.tar elif build_func == "ndk": build_func = ndk.create_shared +elif build_func == "custom": Review comment: I change all conditions to BuildFunc 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] masahi opened a new pull request #7195: [THRUST] Faster multi dimensional argsort by segmented sort
masahi opened a new pull request #7195: URL: https://github.com/apache/tvm/pull/7195 Current implementation of thrust argsort, when given multi dimensional inputs to sort along the inner most axis, is very inefficient: it does `n_iter` calls to thrust sort. See https://github.com/apache/tvm/blob/bad149ed8a555444d813537608ee5cea9e95e97e/src/runtime/contrib/thrust/thrust.cu#L50-L65 When the outer dimension is large, the performance of thrust argsort is far from optimal. In particular, the thrust numbers in shown in https://github.com/apache/tvm/pull/7099 do not reflect the true performance thrust can achieve. This PR replaces `n_iter` calls to thrust argsort with one segmented sort by key. Since thrust doesn't provide API to do sort by key, I used a neat back-to-back stable-sort-by-key trick explained in https://groups.google.com/forum/#!topic/thrust-users/BoLsxO6b4FY. My implementation is a bit more complicated because we need to do segmented sort **by key**, not just segmented sort. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] codeislife99 commented on pull request #7149: Sparse segment sum sqrtn op
codeislife99 commented on pull request #7149: URL: https://github.com/apache/tvm/pull/7149#issuecomment-753730025 @tkonolige @mbrookhart Can I get a re-review on this PR ? I have added the TF Frontend code and some more documentation. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] codeislife99 commented on pull request #7149: Sparse segment sum sqrtn op
codeislife99 commented on pull request #7149: URL: https://github.com/apache/tvm/pull/7149#issuecomment-753729768 Context of these PRs: The goal of adding these sparse ops is to enable a customer to run their recommendation model which is currently getting split into multiple subgraphs because of our non-coverage of this op. I had an offline discussion with the main reviewers but I will also try to summarize the conclusions from it and the comments here: 1. New namespace : Further discussion will be had in a separate thread, after the current (this one included) sparse ops PRs are merged, since there are a few already existing sparse ops without the namespace and if a new namespace is necessary all the current + previous sparse ops will be put in it. 2. Documentation : More documentation has been added 3. High level Approach like XLA: Discussion in a separate thread 4. This operation is TF specific. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] masahi opened a new pull request #7194: [CUBLAS, CUDNN] Support dynamic batch size
masahi opened a new pull request #7194: URL: https://github.com/apache/tvm/pull/7194 Thanks for contributing to TVM! Please refer to guideline https://tvm.apache.org/docs/contribute/ for useful information and tips. After the pull request is submitted, please request code reviews from [Reviewers](https://github.com/apache/incubator-tvm/blob/master/CONTRIBUTORS.md#reviewers) by @ them in the pull request thread. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] codeislife99 commented on a change in pull request #7149: Sparse segment sum sqrtn op
codeislife99 commented on a change in pull request #7149: URL: https://github.com/apache/tvm/pull/7149#discussion_r551089298 ## File path: python/tvm/relay/op/transform.py ## @@ -1320,3 +1320,55 @@ def adv_index(inputs): Output tensor. """ return _make.adv_index(Tuple(inputs)) + + +def sparse_segment_sum(data, indices, segment_ids, num_segments=None): +""" +Compute the sparse segment sum on the indices over the segment_ids Review comment: I will include a link to the TF documentation which will make it clearer. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] codeislife99 commented on a change in pull request #7149: Sparse segment sum sqrtn op
codeislife99 commented on a change in pull request #7149:
URL: https://github.com/apache/tvm/pull/7149#discussion_r551089181
##
File path: src/relay/op/tensor/transform.cc
##
@@ -1553,6 +1553,59 @@ RELAY_REGISTER_OP("meshgrid")
.set_attr("FTVMCompute", MeshgridCompute)
.set_attr("TOpPattern", kInjective);
+TVM_REGISTER_NODE_TYPE(SparseSegmentSumAttrs);
+
+bool SparseSegmentSumRel(const Array& types, int num_inputs, const
Attrs& attrs,
+ const TypeReporter& reporter) {
+ // types: [data, indices, segment_ids, result]
+ ICHECK_EQ(types.size(), 4) << "SparseSegmentSumRel expects 4 types but
provided " << types.size();
+ auto data = types[0].as();
+ auto indices = types[1].as();
+ const auto* param = attrs.as();
+ ICHECK(param != nullptr);
+ Array new_data_shape;
+ new_data_shape.push_back(tvm::max(indices->shape[0], param->num_segments));
+ for (int i = 1; i < static_cast(data->shape.size()); ++i) {
+new_data_shape.push_back(data->shape[i]);
+ }
+ std::vector fields;
+ fields.push_back(TensorType(new_data_shape, data->dtype));
+ fields.push_back(TensorType(Array{1}, tvm::DataType::Int(32)));
+ reporter->Assign(types[3], TupleType(Array(fields)));
+ return true;
+}
+
+Array SparseSegmentSumCompute(const Attrs& attrs, const
Array& inputs,
+ const Type& out_type) {
+ ICHECK_EQ(inputs.size(), 3) << "SparseSegmentSumCompute expects 3 input but
provided "
+ << inputs.size();
+ const auto* param = attrs.as();
+ ICHECK(param != nullptr);
+ return {topi::SparseSegmentSum(inputs[0], inputs[1], inputs[2],
param->num_segments)};
+}
+
+Expr MakeSparseSegmentSum(Expr data, Expr indices, Expr segment_ids, int
num_segments) {
+ auto attrs = make_object();
+ attrs->num_segments = std::move(num_segments);
+ static const Op& op = Op::Get("sparse_segment_sum");
+ return Call(op, {data, indices, segment_ids}, Attrs(attrs), {});
+}
+
+TVM_REGISTER_GLOBAL("relay.op._make.sparse_segment_sum").set_body_typed(MakeSparseSegmentSum);
+
+RELAY_REGISTER_OP("sparse_segment_sum")
+.describe(R"code(Return sparse segment sum of the tensor given segments
+)code" TVM_ADD_FILELINE)
+.set_num_inputs(3)
+.set_attrs_type()
+.add_argument("data", "Tensor", "The first tensor")
+.add_argument("indices", "Tensor", "The second tensor")
+.add_argument("segment_ids", "Tensor", "The third tensor")
Review comment:
Done.
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
[GitHub] [tvm] codeislife99 commented on pull request #7193: Fix ICHECK_NOTNULL in logging.h
codeislife99 commented on pull request #7193: URL: https://github.com/apache/tvm/pull/7193#issuecomment-753719016 @anijain2305 @trevor-m @zhiics Can I get a quick +1 on this ? 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] codeislife99 opened a new pull request #7193: Fix ICHECK_NOTNULL in logging.h
codeislife99 opened a new pull request #7193: URL: https://github.com/apache/tvm/pull/7193 Fix bug in logging.h 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] codeislife99 opened a new pull request #7192: Sparse ops all
codeislife99 opened a new pull request #7192: URL: https://github.com/apache/tvm/pull/7192 Thanks for contributing to TVM! Please refer to guideline https://tvm.apache.org/docs/contribute/ for useful information and tips. After the pull request is submitted, please request code reviews from [Reviewers](https://github.com/apache/incubator-tvm/blob/master/CONTRIBUTORS.md#reviewers) by @ them in the pull request thread. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] sxjscience commented on pull request #7191: [Frontend][MXNet] add _npi_subtract_scalar
sxjscience commented on pull request #7191: URL: https://github.com/apache/tvm/pull/7191#issuecomment-753645201 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 go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] juannzou edited a comment on pull request #6012: [VTA] Move compiler related registry items to vta/build_module.py
juannzou edited a comment on pull request #6012: URL: https://github.com/apache/tvm/pull/6012#issuecomment-752647032 Hi @lhf1997 I had the same issue. Have you tried with [Pynq bitstream 0.0.2](https://github.com/uwsampl/vta-distro/tree/master/bitstreams/pynq/0.0.2)? It worked for me! But then I got another error. I have opened new issue #7182 on that. See also: https://discuss.tvm.apache.org/t/vta-test-fail-while-running-the-2d-convolution-testbench-on-pynq/8789 Hope this helps. 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] juannzou commented on issue #7182: VTA test fail while running the 2D convolution testbench on Pynq
juannzou commented on issue #7182: URL: https://github.com/apache/tvm/issues/7182#issuecomment-753602648 Hi @junrushao1994 I thought it is a bug, since I tried with different settings and still get errors. But I just posted it on the discuss forum, according to your suggestion. However, you seem to have restriction policy there for new users, in terms of how many links or embedded pictures they can include in their posts! So I referenced this post in the forum, for more details. Here is the link to the issue: https://discuss.tvm.apache.org/t/vta-test-fail-while-running-the-2d-convolution-testbench-on-pynq/8789 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[tvm] branch main updated: [Frontend][MXNet] add _npi_subtract_scalar (#7191)
This is an automated email from the ASF dual-hosted git repository.
masahi pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git
The following commit(s) were added to refs/heads/main by this push:
new 86a8504 [Frontend][MXNet] add _npi_subtract_scalar (#7191)
86a8504 is described below
commit 86a8504a7ccb956591ffd1a529f625df7d20b520
Author: insop
AuthorDate: Sun Jan 3 02:46:55 2021 -0800
[Frontend][MXNet] add _npi_subtract_scalar (#7191)
* [Frontend][MXNet] add _npi_subtract_scalar
- add mxnet numpy operator, subtract
- https://github.com/apache/tvm/issues/7186
-
https://mxnet.apache.org/versions/master/api/python/docs/api/np/generated/mxnet.np.subtract.html
* Fix python style using black
---
3rdparty/vta-hw | 2 +-
python/tvm/relay/frontend/mxnet.py | 2 ++
tests/python/frontend/mxnet/test_forward.py | 20
3 files changed, 19 insertions(+), 5 deletions(-)
diff --git a/3rdparty/vta-hw b/3rdparty/vta-hw
index 57db5a7..87ce9ac 16
--- a/3rdparty/vta-hw
+++ b/3rdparty/vta-hw
@@ -1 +1 @@
-Subproject commit 57db5a718c74a788c98120ebbe1230797be698c8
+Subproject commit 87ce9acfae550d1a487746e9d06c2e250076e54c
diff --git a/python/tvm/relay/frontend/mxnet.py
b/python/tvm/relay/frontend/mxnet.py
index f2330c7..1085e90 100644
--- a/python/tvm/relay/frontend/mxnet.py
+++ b/python/tvm/relay/frontend/mxnet.py
@@ -2693,6 +2693,8 @@ _convert_map = {
"_npi_multiply_scalar": _binop_scalar(_op.multiply),
"_npi_add": _rename(_op.add),
"_npi_add_scalar": _binop_scalar(_op.add),
+"_npi_subtract": _rename(_op.subtract),
+"_npi_subtract_scalar": _binop_scalar(_op.subtract),
"_npi_where_rscalar": _mx_npi_where_rscalar,
"_npi_less": _rename(_op.less),
"_npi_less_equal": _mx_compare(_op.less_equal, _rename),
diff --git a/tests/python/frontend/mxnet/test_forward.py
b/tests/python/frontend/mxnet/test_forward.py
index f076a27..d3be8c0 100644
--- a/tests/python/frontend/mxnet/test_forward.py
+++ b/tests/python/frontend/mxnet/test_forward.py
@@ -2062,8 +2062,14 @@ def test_forward_npx_reshape(data_shape, out_shape,
dtype, target, reverse, ctx,
@tvm.testing.parametrize_targets
@pytest.mark.parametrize("kind", ["graph", "vm", "debug"])
def test_forward_npi_binary(data_shape, dtype, target, ctx, kind):
-ref_ops = [mx.np.power, mx.np.multiply, mx.np.add, mx.np.less]
-mx_ops = [mx.sym.np.power, mx.sym.np.multiply, mx.sym.np.add,
mx.sym.np.less]
+ref_ops = [mx.np.power, mx.np.multiply, mx.np.add, mx.np.subtract,
mx.np.less]
+mx_ops = [
+mx.sym.np.power,
+mx.sym.np.multiply,
+mx.sym.np.add,
+mx.sym.np.subtract,
+mx.sym.np.less,
+]
for i in range(len(ref_ops)):
ref_op = ref_ops[i]
mx_op = mx_ops[i]
@@ -2092,8 +2098,14 @@ def test_forward_npi_binary(data_shape, dtype, target,
ctx, kind):
@pytest.mark.parametrize("scalar", [1.0, 2.0, 3.0, 4.0])
@pytest.mark.parametrize("kind", ["graph", "vm", "debug"])
def test_forward_npi_binary_scalar(data_shape, dtype, scalar, target, ctx,
kind):
-ref_ops = [mx.np.power, mx.np.multiply, mx.np.add, mx.np.true_divide]
-mx_ops = [mx.sym.np.power, mx.sym.np.multiply, mx.sym.np.add,
mx.sym.np.true_divide]
+ref_ops = [mx.np.power, mx.np.multiply, mx.np.add, mx.np.subtract,
mx.np.true_divide]
+mx_ops = [
+mx.sym.np.power,
+mx.sym.np.multiply,
+mx.sym.np.add,
+mx.sym.np.subtract,
+mx.sym.np.true_divide,
+]
for i in range(len(ref_ops)):
ref_op = ref_ops[i]
mx_op = mx_ops[i]
[GitHub] [tvm] masahi merged pull request #7191: [Frontend][MXNet] add _npi_subtract_scalar
masahi merged pull request #7191: URL: https://github.com/apache/tvm/pull/7191 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[tvm] branch main updated (6258fae -> 76a9825)
This is an automated email from the ASF dual-hosted git repository. masahi pushed a change to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git. from 6258fae [Fix] Tensor core type issue for dense (#7187) add 76a9825 Remove seemingly invalid SoftPlus (#7189) No new revisions were added by this update. Summary of changes: python/tvm/relay/frontend/onnx.py | 9 - tests/python/frontend/onnx/test_forward.py | 1 - 2 files changed, 10 deletions(-)
[GitHub] [tvm] masahi merged pull request #7189: [FRONTEND][ONNX] Remove seemingly invalid SoftPlus
masahi merged pull request #7189: URL: https://github.com/apache/tvm/pull/7189 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] junrushao1994 commented on pull request #7191: [Frontend][MXNet] add _npi_subtract_scalar
junrushao1994 commented on pull request #7191: URL: https://github.com/apache/tvm/pull/7191#issuecomment-753592836 @sxjscience @eric-haibin-lin Please take a look and let's get it merged 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
