[GitHub] [incubator-tvm] tqchen commented on issue #5440: import vta: Cannot find config in /3rdparty/vta-hw/config/vta_config.json
tqchen commented on issue #5440: URL: https://github.com/apache/incubator-tvm/issues/5440#issuecomment-619479804 Please try the suggested step, pleas also feel free to followup on the https://discuss.tvm.ai/ 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] [incubator-tvm] Menooker commented on a change in pull request #5357: [Relay] enable blocking format in x86 conv2d and fold scale axis
Menooker commented on a change in pull request #5357:
URL: https://github.com/apache/incubator-tvm/pull/5357#discussion_r415208733
##
File path: src/relay/transforms/fold_scale_axis.cc
##
@@ -314,6 +318,42 @@ class ForwardPrep : private ExprVisitor {
}
};
+static bool IsIntInArray(const Array& axis, int v) {
+ for (size_t i = 0; i < axis.size(); i++) {
+if (axis[i] == v)
+ return true;
+ }
+ return false;
+}
+
+static Expr ReshapeToMatchAxis(Expr scale, const Array& shape,
+ const Array& axis) {
+ Array arr;
+ for (size_t i = 0; i < shape.size(); i++) {
+if (IsIntInArray(axis, i)) {
+ auto node = shape[i].as();
+ if (!node) {
+// if the shape is not a constant, use normal transform
+return Expr();
Review comment:
I have changed `CHECK(scale.defined())` now. It now fallbacks to a
"optimization failure" rather than a compilation error.
For ForwardRewrite, if the shape is not constant, the rewriter will return
`Expr()`.
For BackwardRewrite, if the shape is not constant, the rewriter will return
`transformer->NormalCallTransform(call.operator->())`.
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] [incubator-tvm] Menooker commented on a change in pull request #5357: [Relay] enable blocking format in x86 conv2d and fold scale axis
Menooker commented on a change in pull request #5357:
URL: https://github.com/apache/incubator-tvm/pull/5357#discussion_r415207561
##
File path: src/relay/transforms/fold_scale_axis.cc
##
@@ -39,6 +39,10 @@ namespace relay {
*
* Use namespace to reduce potential naming conflict.
*/
+
+extern Expr MakeReshape(Expr data,
+ Array newshape);
Review comment:
Would you plz suggest a `.h` where we can move this declaration to? The
function is defined in `relay/transform.cc`. But it is not declared anywhere in
a header file.
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] [incubator-tvm] Menooker commented on a change in pull request #5357: [Relay] enable blocking format in x86 conv2d and fold scale axis
Menooker commented on a change in pull request #5357:
URL: https://github.com/apache/incubator-tvm/pull/5357#discussion_r415206359
##
File path: src/relay/transforms/fold_scale_axis.cc
##
@@ -518,13 +564,30 @@ Expr Conv2DForwardRewrite(const Call& ref_call,
// match the ic_axis
if (is_depthwise_conv2d) {
-Expr scale = ExpandBiasToMatchAxis(
-sdata->scale, kernel_layout.ndim(), {big_oc_axis});
-weight = Multiply(weight, scale);
+if (is_simple) {
+ Expr scale = ExpandBiasToMatchAxis(
+ sdata->scale, kernel_layout.ndim(), {big_ko_axis});
+ weight = Multiply(weight, scale);
+} else {
+ weight = Multiply(weight, ReshapeToMatchAxis(sdata->scale,
+ weight->type_as()->shape,
+ {big_ko_axis, small_ko_axis}));
+ if (!weight.defined())
+return Expr();
Review comment:
If I am understanding correctly, line 166~175 in
`src/relay/transforms/forward_rewrite.cc` will take care of `Expr()`
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] [incubator-tvm] Menooker commented on a change in pull request #5357: [Relay] enable blocking format in x86 conv2d and fold scale axis
Menooker commented on a change in pull request #5357:
URL: https://github.com/apache/incubator-tvm/pull/5357#discussion_r415204685
##
File path: src/relay/transforms/fold_scale_axis.cc
##
@@ -518,13 +564,30 @@ Expr Conv2DForwardRewrite(const Call& ref_call,
// match the ic_axis
if (is_depthwise_conv2d) {
-Expr scale = ExpandBiasToMatchAxis(
-sdata->scale, kernel_layout.ndim(), {big_oc_axis});
-weight = Multiply(weight, scale);
+if (is_simple) {
+ Expr scale = ExpandBiasToMatchAxis(
+ sdata->scale, kernel_layout.ndim(), {big_ko_axis});
+ weight = Multiply(weight, scale);
+} else {
+ weight = Multiply(weight, ReshapeToMatchAxis(sdata->scale,
+ weight->type_as()->shape,
+ {big_ko_axis, small_ko_axis}));
+ if (!weight.defined())
+return Expr();
Review comment:
I can see in line 542 and 543 in the same function:
```
if (sdata == nullptr) return Expr();
if (sweight != nullptr) return Expr();
```
Looks like returning `Expr()` just mean that the optimization fails?
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] [incubator-tvm] FrozenGene commented on pull request #5444: [KERAS]Embedding layer
FrozenGene commented on pull request #5444: URL: https://github.com/apache/incubator-tvm/pull/5444#issuecomment-619472461 Thanks @siju-samuel 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
[incubator-tvm] branch master updated: [KERAS]Embedding layer (#5444)
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 1014fef [KERAS]Embedding layer (#5444)
1014fef is described below
commit 1014fefa54b5f0a359501b6d19ea3b5a52d6dca6
Author: Samuel
AuthorDate: Sun Apr 26 08:28:02 2020 +0530
[KERAS]Embedding layer (#5444)
---
python/tvm/relay/frontend/keras.py | 10 +-
tests/python/frontend/keras/test_forward.py | 20 +++-
2 files changed, 28 insertions(+), 2 deletions(-)
diff --git a/python/tvm/relay/frontend/keras.py
b/python/tvm/relay/frontend/keras.py
index bf91bc1..43065be 100644
--- a/python/tvm/relay/frontend/keras.py
+++ b/python/tvm/relay/frontend/keras.py
@@ -207,6 +207,14 @@ def _convert_permute(inexpr, keras_layer, _):
return _op.transpose(inexpr, axes=(0,) + keras_layer.dims)
+def _convert_embedding(inexpr, keras_layer, etab):
+indices = inexpr
+weightList = keras_layer.get_weights()
+weight = etab.new_const(weightList[0])
+out = _op.take(weight, indices.astype('int32'), axis=0)
+
+return out
+
def _convert_dense(inexpr, keras_layer, etab):
weightList = keras_layer.get_weights()
weight = etab.new_const(weightList[0].transpose([1, 0]))
@@ -893,7 +901,7 @@ _convert_map = {
'Maximum' : _convert_merge,
'Dot' : _convert_merge,
'Permute' : _convert_permute,
-# 'Embedding' : _convert_embedding,
+'Embedding': _convert_embedding,
# 'RepeatVector' : _convert_repeat_vector,
'InputLayer' : _default_skip,
diff --git a/tests/python/frontend/keras/test_forward.py
b/tests/python/frontend/keras/test_forward.py
index b764137..b4a1816 100644
--- a/tests/python/frontend/keras/test_forward.py
+++ b/tests/python/frontend/keras/test_forward.py
@@ -466,6 +466,24 @@ class TestKeras:
keras_model = keras.models.Model(data, x)
verify_keras_frontend(keras_model, layout='NDHWC')
+
+def test_forward_embedding(self, keras):
+data = keras.layers.Input(shape=(2, 4), dtype="int32")
+x = keras.layers.Embedding(10, 3)(data)
+keras_model = keras.models.Model(data, x)
+verify_keras_frontend(keras_model, need_transpose=False)
+
+data = keras.layers.Input(shape=(2, 3, 4), dtype="int32")
+x = keras.layers.Embedding(4, 5)(data)
+keras_model = keras.models.Model(data, x)
+verify_keras_frontend(keras_model, need_transpose=False)
+
+data = keras.layers.Input(shape=(6, 2, 3, 4), dtype="int32")
+x = keras.layers.Embedding(4, 5)(data)
+keras_model = keras.models.Model(data, x)
+verify_keras_frontend(keras_model, need_transpose=False)
+
+
if __name__ == '__main__':
for k in [keras, tf_keras]:
sut = TestKeras()
@@ -497,4 +515,4 @@ if __name__ == '__main__':
sut.test_forward_pool3d(keras=k)
sut.test_forward_upsample3d(keras=k)
sut.test_forward_zero_padding3d(keras=k)
-
+sut.test_forward_embedding(keras=k)
[GitHub] [incubator-tvm] lixiaoquan commented on a change in pull request #5429: [RELAY][TF] Support symbolic newshape for Reshape
lixiaoquan commented on a change in pull request #5429:
URL: https://github.com/apache/incubator-tvm/pull/5429#discussion_r415197766
##
File path: include/tvm/relay/op_attr_types.h
##
@@ -81,10 +81,16 @@ using TOpIsStateful = bool;
*/
using TNonComputational = bool;
+enum ShapeDependantKind {
+ kShapeDependantShape = 0,
+ kShapeDependantData = 1,
+ kShapeDependantBoth = 2,
Review comment:
Thanks for you comment.
I tried two ways to use data tensor's shape directly.
First, I passed data tensor's shape to _reshape_shape_func but I got this
error
```
ValueError: All indices are supposed to be constants
```
Second, I tried to construct a tensor by tvm.te.placeholder based on data
tensor's shape, but it doesn't work because it can't be passed to tvm.lower
called later.
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] [incubator-tvm] lixiaoquan commented on a change in pull request #5429: [RELAY][TF] Support symbolic newshape for Reshape
lixiaoquan commented on a change in pull request #5429:
URL: https://github.com/apache/incubator-tvm/pull/5429#discussion_r415196802
##
File path: include/tvm/relay/op_attr_types.h
##
@@ -81,10 +81,16 @@ using TOpIsStateful = bool;
*/
using TNonComputational = bool;
+enum ShapeDependantKind {
+ kShapeDependantShape = 0,
+ kShapeDependantData = 1,
+ kShapeDependantBoth = 2,
Review comment:
Thanks for your comment
I checked #4312, and tried to use data.shape the same ways as
_strided_slice_shape_func(), but I got this error from hybrid script parser
```
ValueError: All indices are supposed to be constants
```
it seems _strided_slice_shape_func only uses const index when accessing
data.shape, but _reshape_shape_func uses variables as indices to access
data_shape
I'd appreciate it if you can give me some hint about how to solve 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] [incubator-tvm] lixiaoquan commented on a change in pull request #5429: [RELAY][TF] Support symbolic newshape for Reshape
lixiaoquan commented on a change in pull request #5429:
URL: https://github.com/apache/incubator-tvm/pull/5429#discussion_r415196802
##
File path: include/tvm/relay/op_attr_types.h
##
@@ -81,10 +81,16 @@ using TOpIsStateful = bool;
*/
using TNonComputational = bool;
+enum ShapeDependantKind {
+ kShapeDependantShape = 0,
+ kShapeDependantData = 1,
+ kShapeDependantBoth = 2,
Review comment:
I checked #4312, and tried to use data.shape the same ways as
_strided_slice_shape_func(), but I got this error from hybrid script parser
```
ValueError: All indices are supposed to be constants
```
it seems _strided_slice_shape_func only uses const index when accessing
data.shape, but _reshape_shape_func uses variables as indices to access
data_shape
I'd appreciate it if you can give me some hint about how to solve 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] [incubator-tvm] libaihong commented on pull request #5428: [CODEGEN][CUDA] Fix a bug when vectorized load&store was involved for…
libaihong commented on pull request #5428: URL: https://github.com/apache/incubator-tvm/pull/5428#issuecomment-619465094 > > > Could you please add a few tests.Do you mean loading uchar2x4? We could load/store them as uint16_t x 4. I do not see similar code in PrintType below: > > https://github.com/apache/incubator-tvm/blob/master/src/target/source/codegen_cuda.cc#L186 I've added some unit test. > > > Could you please add a few tests.Do you mean loading uchar2x4? We could load/store them as uint16_t x 4. I do not see similar code in PrintType below: > > https://github.com/apache/incubator-tvm/blob/master/src/target/source/codegen_cuda.cc#L186 I've added some unit test. This modification only add code for char2 when the lanes is 2. And when lanes is 4, it will use the old logic and will not loading uchar4. 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] [incubator-tvm] yzhliu commented on a change in pull request #5357: [Relay] enable blocking format in x86 conv2d and fold scale axis
yzhliu commented on a change in pull request #5357:
URL: https://github.com/apache/incubator-tvm/pull/5357#discussion_r415164446
##
File path: src/relay/transforms/fold_scale_axis.cc
##
@@ -314,6 +318,42 @@ class ForwardPrep : private ExprVisitor {
}
};
+static bool IsIntInArray(const Array& axis, int v) {
+ for (size_t i = 0; i < axis.size(); i++) {
+if (axis[i] == v)
+ return true;
+ }
+ return false;
+}
+
+static Expr ReshapeToMatchAxis(Expr scale, const Array& shape,
+ const Array& axis) {
+ Array arr;
+ for (size_t i = 0; i < shape.size(); i++) {
+if (IsIntInArray(axis, i)) {
+ auto node = shape[i].as();
+ if (!node) {
+// if the shape is not a constant, use normal transform
+return Expr();
Review comment:
this will cause failure as you later do `CHECK(scale.defined());` ?
##
File path: src/relay/transforms/fold_scale_axis.cc
##
@@ -39,6 +39,10 @@ namespace relay {
*
* Use namespace to reduce potential naming conflict.
*/
+
+extern Expr MakeReshape(Expr data,
+ Array newshape);
Review comment:
can we move it to .h?
##
File path: src/relay/transforms/fold_scale_axis.cc
##
@@ -518,13 +564,30 @@ Expr Conv2DForwardRewrite(const Call& ref_call,
// match the ic_axis
if (is_depthwise_conv2d) {
-Expr scale = ExpandBiasToMatchAxis(
-sdata->scale, kernel_layout.ndim(), {big_oc_axis});
-weight = Multiply(weight, scale);
+if (is_simple) {
+ Expr scale = ExpandBiasToMatchAxis(
+ sdata->scale, kernel_layout.ndim(), {big_ko_axis});
+ weight = Multiply(weight, scale);
+} else {
+ weight = Multiply(weight, ReshapeToMatchAxis(sdata->scale,
+ weight->type_as()->shape,
+ {big_ko_axis, small_ko_axis}));
+ if (!weight.defined())
+return Expr();
Review comment:
would this cause problem?
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] [incubator-tvm] icemelon9 commented on a change in pull request #5429: [RELAY][TF] Support symbolic newshape for Reshape
icemelon9 commented on a change in pull request #5429:
URL: https://github.com/apache/incubator-tvm/pull/5429#discussion_r415137698
##
File path: include/tvm/relay/op_attr_types.h
##
@@ -81,10 +81,16 @@ using TOpIsStateful = bool;
*/
using TNonComputational = bool;
+enum ShapeDependantKind {
+ kShapeDependantShape = 0,
+ kShapeDependantData = 1,
+ kShapeDependantBoth = 2,
Review comment:
No need to have `kShapeDependantBoth`. When you provide the data tensor
as input to the shape function, you can retrieve its shape directly. So it's
unnecessary to provide both shape and data tensor.
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] [incubator-tvm] kevinthesun commented on a change in pull request #5429: [RELAY][TF] Support symbolic newshape for Reshape
kevinthesun commented on a change in pull request #5429:
URL: https://github.com/apache/incubator-tvm/pull/5429#discussion_r415134966
##
File path: include/tvm/relay/op_attr_types.h
##
@@ -81,10 +81,16 @@ using TOpIsStateful = bool;
*/
using TNonComputational = bool;
+enum ShapeDependantKind {
+ kShapeDependantShape = 0,
+ kShapeDependantData = 1,
+ kShapeDependantBoth = 2,
Review comment:
Also this change involves a lot of changes in compile engine as well.
AFAIK, dynamic stridedslice doesn't require these changes. Dynamic reshape
should not be that different?
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] [incubator-tvm] kevinthesun commented on a change in pull request #5429: [RELAY][TF] Support symbolic newshape for Reshape
kevinthesun commented on a change in pull request #5429:
URL: https://github.com/apache/incubator-tvm/pull/5429#discussion_r415134500
##
File path: include/tvm/relay/op_attr_types.h
##
@@ -81,10 +81,16 @@ using TOpIsStateful = bool;
*/
using TNonComputational = bool;
+enum ShapeDependantKind {
+ kShapeDependantShape = 0,
+ kShapeDependantData = 1,
+ kShapeDependantBoth = 2,
Review comment:
A bit curious about whether we need kShapeDependantBoth. Currently we
can support data dependent shape function, and reshape with symbolic newshape
should also be data dependent.
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] [incubator-tvm] antinucleon commented on issue #5427: [Relay] Lack complex tests for parser
antinucleon commented on issue #5427:
URL: https://github.com/apache/incubator-tvm/issues/5427#issuecomment-619423150
Before the parser is fixed, in case someone needs a parser in a hurry, here
is my emergency use parser:
```
import re
import random
import tvm
from tvm import auto_scheduler
from tvm import relay
class ReRelayParser(object):
def __init__(self, expr_text, param_dict):
self.memo = {}
self.expr_text = expr_text
self.param_dict = param_dict
def _extract_args(self):
ARG_LIST = re.compile(r'fn\s\((.*?)\)\s->')
ARGS =
re.compile(r'\%([a-zA-Z0-9_]+):\sTensor\[\(([0-9,\s]+)\),\s([a-z0-9]+)\]')
arg_list = ARG_LIST.findall(self.expr_text)[0]
args = ARGS.findall(arg_list)
ret = []
for name, ss, dtype in args:
cmd = "{name} = relay.var('{name}', shape=[{shape}],
dtype='{dtype}')".format(name=name, shape=ss, dtype=dtype)
ret.append(cmd)
self.memo["%" + name] = name
return ret
def _extract_body(self):
ret = []
STMT =
re.compile(r'(\%\d+)\s=\s([a-zA-Z0-9._]+)\(([a-zA-Z\%0-9,\s_=\[\]"]+)\)')
LAST_STMT =
re.compile(r'([a-zA-Z0-9._]+)\(([a-zA-Z\%0-9,\s_=\[\]"]+)\)')
def random_name(length=8):
name = ""
for i in range(length):
name += chr(random.randint(97, 122))
return name
def process_args(args):
VAR_ARGS = re.compile(r'(%[0-9a-zA-z_]+)')
var_arg = VAR_ARGS.findall(args)
var_len = len(var_arg)
tmp = args.split(",")
return var_arg, ",".join(tmp[var_len:])
tmp = STMT.findall(self.expr_text)
for ss in tmp:
output = ss[0]
op = ss[1]
args = ss[2]
self.memo[output] = random_name()
var_args, attrs = process_args(args)
cmd = "{name} = relay.{op}({var_args}, {attrs})".format(
name=self.memo[output],
op=op,
var_args=", ".join([self.memo[x] for x in var_args]),
attrs=attrs
)
ret.append(cmd)
op, args = LAST_STMT.findall(self.expr_text)[-1]
var_args, attrs = process_args(args)
cmd = "output_expr = relay.{op}({var_args}, {attrs})".format(
op=op,
var_args=", ".join([self.memo[x] for x in var_args]),
attrs=attrs
)
ret.append(cmd)
return ret
def _convert_parmas(self):
ret = {}
for key in self.param_dict.keys():
new_key = "v" + key
ret[new_key] = self.param_dict[key]
return ret
def convert(self):
cmds = []
cmds.extend(self._extract_args())
cmds.extend(self._extract_body())
new_params = self._convert_parmas()
l = {}
exec("\n".join(cmds), None, l)
output_expr = l["output_expr"]
func = relay.Function(relay.analysis.free_vars(output_expr),
output_expr)
# mod = tvm.ir.IRModule.from_expr(output_expr)
return func, new_params
if __name__ == "__main__":
from PIL import Image
from matplotlib import pyplot as plt
import numpy as np
from tvm.contrib.download import download_testdata
image_url =
'https://github.com/dmlc/mxnet.js/blob/master/data/cat.png?raw=true'
image_path = download_testdata(image_url, 'cat.png', module='data')
resized_image = Image.open(image_path).resize((224, 224))
#plt.imshow(resized_image)
#plt.show()
image_data = np.asarray(resized_image).astype("float32")
# Add a dimension to the image so that we have NHWC format layout
image_data = np.expand_dims(image_data, axis=0)
synset_url = ''.join(['https://gist.githubusercontent.com/zhreshold/',
'4d0b62f3d01426887599d4f7ede23ee5/raw/',
'596b27d23537e5a1b5751d2b0481ef172f58b539/',
'imagenet1000_clsid_to_human.txt'])
synset_name = 'imagenet1000_clsid_to_human.txt'
synset_path = download_testdata(synset_url, synset_name, module='data')
with open(synset_path) as f:
synset = eval(f.read())
# Preprocess image as described here:
#
https://github.com/tensorflow/models/blob/edb6ed22a801665946c63d650ab9a0b23d98e1b1/research/slim/preprocessing/inception_preprocessing.py#L243
#image_data[:, :, :, 0] = 2.0 / 255.0 * image_data[:, :, :, 0] - 1
#image_data[:, :, :, 1] = 2.0 / 255.0 * image_data[:, :, :, 1] - 1
#image_data[:, :, :, 2] = 2.
[GitHub] [incubator-tvm] mbrookhart commented on pull request #5441: Add TopK to ONNX Frontend
mbrookhart commented on pull request #5441: URL: https://github.com/apache/incubator-tvm/pull/5441#issuecomment-619418119 Thank you! 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] [incubator-tvm] siju-samuel opened a new pull request #5444: [KERAS]Embedding layer
siju-samuel opened a new pull request #5444: URL: https://github.com/apache/incubator-tvm/pull/5444 @FrozenGene Please help to review the embedding layer op in Keras. 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] [incubator-tvm] woniuasd commented on issue #5133: [Torch] A list of missing op conversion in need of help
woniuasd commented on issue #5133: URL: https://github.com/apache/incubator-tvm/issues/5133#issuecomment-619408318 @siju-samuel You mean i can use embedded instead of aten::index_put_ ？ 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
[incubator-tvm] branch master updated: [RELAY] Move frontend utils (#5345)
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 3f47b32 [RELAY] Move frontend utils (#5345) 3f47b32 is described below commit 3f47b32774dfcbe8f51f0ab0d86e09535904cb8d Author: mbaret <55580676+mba...@users.noreply.github.com> AuthorDate: Sat Apr 25 17:18:30 2020 +0100 [RELAY] Move frontend utils (#5345) * [RELAY] Move frontend utils The util file currently under frontend is used from outside of frontend (in qnn/op/legalizations). This suggests that the file should be pushed up to a higher level. The benefit from this change is that importing qnn no longer also imports all the frontends. * Inline get_scalar_from_constant Change-Id: I1cc64e9ecb0eadb6ac0f7b62e6ea174644af4ad4 * Remove util.py from Relay Change-Id: If9cd7cf3fc0bd1861a3a9b5604f338e084d8db96 * Shorten functions Change-Id: Ieb537d82e6ee52421ff05a90cd00a03679ffebf2 * Line length Change-Id: I1d216b7e73a060c4f118f5da50ce58b18eba907f --- python/tvm/relay/frontend/tflite.py | 12 +++- python/tvm/relay/frontend/util.py| 33 python/tvm/relay/qnn/op/legalizations.py | 11 ++- 3 files changed, 21 insertions(+), 35 deletions(-) diff --git a/python/tvm/relay/frontend/tflite.py b/python/tvm/relay/frontend/tflite.py index 275d0ce..bba7d3b 100644 --- a/python/tvm/relay/frontend/tflite.py +++ b/python/tvm/relay/frontend/tflite.py @@ -29,7 +29,6 @@ from .. import function as _function from .. import op as _op from .. import qnn as _qnn from ... import nd as _nd -from .util import get_scalar_from_constant from .common import ExprTable from .common import infer_shape as _infer_shape @@ -2281,6 +2280,17 @@ class OperatorConverter(object): def has_expr(self, input_tensor_idx): return self.exp_tab.has_expr(get_tensor_name(self.subgraph, input_tensor_idx)) + +def get_scalar_from_constant(expr): +""" Returns scalar value from Relay constant scalar. """ +assert isinstance(expr, _expr.Constant) and not expr.data.shape, \ +"Expr is not a constant scalar." +value = expr.data.asnumpy() +assert value.dtype == np.dtype(np.int32) or value.dtype == np.dtype(np.float32), \ +"value must be float32/int32" +return np.asscalar(value) + + def build_str_map(obj): """Build string map of TFLite enum int value diff --git a/python/tvm/relay/frontend/util.py b/python/tvm/relay/frontend/util.py deleted file mode 100644 index a7f89a3..000 --- a/python/tvm/relay/frontend/util.py +++ /dev/null @@ -1,33 +0,0 @@ -# Licensed to the Apache Software Foundation (ASF) under one -# or more contributor license agreements. See the NOTICE file -# distributed with this work for additional information -# regarding copyright ownership. The ASF licenses this file -# to you under the Apache License, Version 2.0 (the -# "License"); you may not use this file except in compliance -# with the License. You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, -# software distributed under the License is distributed on an -# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -# KIND, either express or implied. See the License for the -# specific language governing permissions and limitations -# under the License. -# pylint: disable=wildcard-import, redefined-builtin, invalid-name -""" Utility functions that are used across many directories. """ -from __future__ import absolute_import -import numpy as np -from .. import expr as _expr - -def get_scalar_from_constant(expr): -""" Returns scalar value from Relay constant scalar. """ -assert isinstance(expr, _expr.Constant) and not expr.data.shape, \ -"Expr is not a constant scalar." -value = expr.data.asnumpy() -if value.dtype == np.dtype(np.int32): -return int(value) -if value.dtype == np.dtype(np.float32): -return float(value) -assert False, "Constant expr must be float32/int32" -return None # To suppress pylint diff --git a/python/tvm/relay/qnn/op/legalizations.py b/python/tvm/relay/qnn/op/legalizations.py index b1c1909..c96a730 100644 --- a/python/tvm/relay/qnn/op/legalizations.py +++ b/python/tvm/relay/qnn/op/legalizations.py @@ -20,8 +20,8 @@ from __future__ import absolute_import import tvm from tvm import relay +import numpy as np from .. import op as reg -from ...frontend.util import get_scalar_from_constant # # Register the functions for different operators. @@ -54,6 +54,15 @@ def qnn_dense_legalize(attrs, inputs, types): # Helper functions. ### +def get_scalar_from_con
[GitHub] [incubator-tvm] HK017 commented on issue #5443: How to implement a demo of TVM C++ object-detection?
HK017 commented on issue #5443: URL: https://github.com/apache/incubator-tvm/issues/5443#issuecomment-619376478 > Suggest asking this question on the discuss forum: https://discuss.tvm.ai/ Because no one replied to me, I came here for help on the discuss forum: https://discuss.tvm.ai/ 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] [incubator-tvm] FrozenGene commented on issue #5443: How to implement a demo of TVM C++ object-detection?
FrozenGene commented on issue #5443: URL: https://github.com/apache/incubator-tvm/issues/5443#issuecomment-619351941 Suggest asking this question on the discuss forum: https://discuss.tvm.ai/ 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] [incubator-tvm] HK017 opened a new issue #5443: How to implement a demo of TVM C++ object-detection?
HK017 opened a new issue #5443: URL: https://github.com/apache/incubator-tvm/issues/5443 hi all developer， how to transform tvm::run_time::NDArray data to c++ data so that I can process the data better，for example， For yolov3 (Darknet version) TVM C + + code, how to handle the output of get_output? If you see the problem, please help me to solve it, I will be very grateful！！！ Thanks again 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
[incubator-tvm] branch master updated: [CodeGen] Cleanup generated code (#5424)
This is an automated email from the ASF dual-hosted git repository.
wuwei 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 952def5 [CodeGen] Cleanup generated code (#5424)
952def5 is described below
commit 952def53da51e6cf17c5dbf50b92e193622ca695
Author: Wei Pan <60017475+wpan1...@users.noreply.github.com>
AuthorDate: Sat Apr 25 01:20:27 2020 -0700
[CodeGen] Cleanup generated code (#5424)
- remove unnecessary white spaces from storage kind
- do not start a new scope for vectorization as temporary
variables are alll uniquely generated.
The above two changes make vectorized code much cleaner.
Signed-off-by: Wei Pan
---
src/target/source/codegen_c.cc | 10 +-
src/target/source/codegen_c.h | 23 ---
src/target/source/codegen_cuda.cc | 10 +-
src/target/source/codegen_metal.cc | 7 +++
src/target/source/codegen_opencl.cc | 5 ++---
5 files changed, 7 insertions(+), 48 deletions(-)
diff --git a/src/target/source/codegen_c.cc b/src/target/source/codegen_c.cc
index 84604b8..6461908 100644
--- a/src/target/source/codegen_c.cc
+++ b/src/target/source/codegen_c.cc
@@ -94,7 +94,6 @@ void CodeGenC::AddFunction(const PrimFunc& f) {
auto it = alloc_storage_scope_.find(v.get());
if (it != alloc_storage_scope_.end()) {
PrintStorageScope(it->second, stream);
-stream << ' ';
}
PrintType(GetType(v), stream);
@@ -179,7 +178,6 @@ std::string CodeGenC::GetBufferRef(
if (!scope.empty() && IsScopePartOfType()) {
PrintStorageScope(scope, os);
}
- os << ' ';
PrintType(t, os);
os << "*)" << vid << ')';
} else {
@@ -213,7 +211,6 @@ std::string CodeGenC::GetBufferRef(
if (!scope.empty() && IsScopePartOfType()) {
PrintStorageScope(scope, os);
}
-os << ' ';
PrintType(t, os);
os << "*)(";
if (!HandleTypeMatch(buffer, t.element_of())) {
@@ -221,7 +218,6 @@ std::string CodeGenC::GetBufferRef(
if (!scope.empty() && IsScopePartOfType()) {
PrintStorageScope(scope, os);
}
- os << ' ';
PrintType(t.element_of(), os);
os << "*)";
}
@@ -681,7 +677,6 @@ void CodeGenC::VisitExpr_(const LoadNode* op, std::ostream&
os) { // NOLINT(*)
auto it = alloc_storage_scope_.find(op->buffer_var.get());
if (it != alloc_storage_scope_.end()) {
PrintStorageScope(it->second, value_temp);
- value_temp << ' ';
}
}
PrintType(elem_type, value_temp);
@@ -731,7 +726,6 @@ void CodeGenC::VisitStmt_(const StoreNode* op) {
auto it = alloc_storage_scope_.find(op->buffer_var.get());
if (it != alloc_storage_scope_.end()) {
PrintStorageScope(it->second, stream);
- stream << ' ';
}
}
PrintType(elem_type, stream);
@@ -823,10 +817,8 @@ void CodeGenC::VisitStmt_(const AllocateNode* op) {
const VarNode* buffer = op->buffer_var.as();
std::string scope = alloc_storage_scope_.at(buffer);
PrintStorageScope(scope, stream);
-stream << ' ';
PrintType(op->dtype, stream);
-stream << ' '<< vid << '['
- << constant_size << "];\n";
+stream << ' ' << vid << '[' << constant_size << "];\n";
RegisterHandleType(op->buffer_var.get(), op->dtype);
this->PrintStmt(op->body);
diff --git a/src/target/source/codegen_c.h b/src/target/source/codegen_c.h
index db655be..4fb4b7e 100644
--- a/src/target/source/codegen_c.h
+++ b/src/target/source/codegen_c.h
@@ -257,29 +257,6 @@ class CodeGenC :
/*! \brief the data type of allocated buffers */
std::unordered_map handle_data_type_;
- /*!
- * \brief A RAII utility class for emitting code in a scoped region.
- */
- class EnterScopeRAII {
-// The codegen context.
-CodeGenC* cg;
-
-// The new scope level.
-int scope;
-
- public:
-explicit EnterScopeRAII(CodeGenC* cg) : cg(cg) {
- cg->PrintIndent();
- cg->stream << "{\n";
- scope = cg->BeginScope();
-}
-~EnterScopeRAII() {
- cg->EndScope(scope);
- cg->PrintIndent();
- cg->stream << "}\n";
-}
- };
-
private:
/*! \brief whether to print in SSA form */
bool print_ssa_form_{false};
diff --git a/src/target/source/codegen_cuda.cc
b/src/target/source/codegen_cuda.cc
index 02b5b41..56d7162 100644
--- a/src/target/source/codegen_cuda.cc
+++ b/src/target/source/codegen_cuda.cc
@@ -242,8 +242,6 @@ void CodeGenCUDA::PrintVecBinaryOp(
this->PrintType(t, stream);
stream << ' ' << sret << ";\n";
{
-EnterScopeRAII scope(this);
-
// Unpack into individual ops.
std::string vlhs = SSAGetID(PrintExpr(lhs), lhs.dtype());
std::string vrhs = SSAGetID(PrintExpr(rhs), rhs.dtype());
@@ -350,7 +348,7 @@
