Hello,
I was looking at the Arm EthosU integration in TVM and [noticed that there was
a new conv2d Relay operator
defined](https://github.com/apache/tvm/blob/main/python/tvm/relay/backend/contrib/ethosu/op/convolution.py#L185).
Obviously this operator is only legal/valid for offloading onto t
But it seems that we should not consider the `min(1, OH - i) = 1`, but directly:
for(i=0; ihttps://discuss.tvm.apache.org/t/a-failed-example-of-using-compute-at-based-on-tvmscript/11489/6)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these
The analyzer doesn't seem to be able to prove the following equality:
```python
min(1, OH - i) = 1
```
because the range of `OH` is not known
---
[Visit
Topic](https://discuss.tvm.apache.org/t/a-failed-example-of-using-compute-at-based-on-tvmscript/11489/5)
to respond.
You are receiving
Yeah，I think maybe I should quit the road of windows now
---
[Visit
Topic](https://discuss.tvm.apache.org/t/check-failed-reporter-asserteq-data-shape-data-shape-size-1-weight-shape-1-is-false-denserel-input-dimension-doesnt-match-data-shape-1-512-weight-shape-512-1000/11274/17)
to respond.
Thank you for your reply, this is the way I want.
According to your suggestion, I did an experiment and changed the constant 128
to the variable OH,OW.
@T.prim_func
def compute_at_call_extern(a: T.handle, c: T.handle) -> None:
T.func_attr({"global_symbol": "main", "tir.noal
Oh you were using windows this whole time???
Yeah I think it might be a little tricky, it has the least amount of testing
and use I believe.
---
[Visit
Topic](https://discuss.tvm.apache.org/t/check-failed-reporter-asserteq-data-shape-data-shape-size-1-weight-shape-1-is-false-denserel-inp
I was able to run successfully on torch1.8.0, but reported a new error
"RuntimeError: LLVM version is not available, please check if you build with
LLVM ", maybe there is a problem with TVM and LLVM on Windows, I am ready to
give up and use Ubuntu in the future. There are too many bugs running
I got an error running on torch1.9.0
---
[Visit
Topic](https://discuss.tvm.apache.org/t/check-failed-reporter-asserteq-data-shape-data-shape-size-1-weight-shape-1-is-false-denserel-input-dimension-doesnt-match-data-shape-1-512-weight-shape-512-1000/11274/14)
to respond.
You are receiving
Hi, thanks for your reply.
the tensorflow model is from
https://zenodo.org/record/3345892/files/tf_ssd_resnet34_22.1.zip?download=1
in page :
https://github.com/mlcommons/inference/tree/master/vision/classification_and_detection
actually, the script is simple like below:
with tf.gfile.
You mean combine the two kernels? But actually I want to cut the if statements.
---
[Visit
Topic](https://discuss.tvm.apache.org/t/ops-become-slow-when-using-te-var/11486/5)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
Thank you your reply. Yeah,my next plan is using auto scheduling, I just do not
understand why quantize model is so slow.
---
[Visit
Topic](https://discuss.tvm.apache.org/t/using-tvm-quantize-model-is-too-slower-than-not-quantize/11487/3)
to respond.
You are receiving this because you en
To get better performance, you can try auto tuning or auto scheduling
You may find these tutorials helpful
https://tvm.apache.org/docs/how_to/tune_with_autotvm/tune_relay_cuda.html
https://tvm.apache.org/docs/how_to/tune_with_autoscheduler/tune_network_cuda.html?highlight=tune%20relay
---
[V
I see. I feel like your intention is to move block "C" under certain loops
above block "B". Is that correct. If so, you may use `reverse_compute_at` in
your particular case.
try this:
```python
def test_compute_at2():
sch = tir.Schedule(compute_at_call_extern, debug_mask="all")
print(
We don't support opaque access to buffers
---
[Visit
Topic](https://discuss.tvm.apache.org/t/a-failed-example-of-using-compute-at-based-on-tvmscript/11489/2)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
here](https://
Hi, everyone!
I'm struggling with deploying an UNET-like model on the Hexagon 780 DSP.
Does the newest version support the Hexagon 780?
Anyone has succussed running a model on theHexagon 780?
---
[Visit
Topic](https://discuss.tvm.apache.org/t/utilizing-hexagon-780-dsp-of-snapdragon-88
I am a TensorIR/TVMScript beginner, I did an experiment, triggered some errors,
how to deal with it? Thanks.
@T.prim_func
def compute_at_call_exterm(a: T.handle, c: T.handle) -> None:
T.func_attr({"global_symbol": "main", "tir.noalias": True})
A = T.match_buffer(a, (128,
**My target is "cuda", and I use the first way to quantize
model**(https://tvm.apache.org/docs/how_to/deploy_models/deploy_quantized.html#sphx-glr-how-to-deploy-models-deploy-quantized-py)
**and then inference speed is very slow, Why
Just inline the one stage into the other one?
EDIT:
wait your if statements require variables which are not defined (blockIdx.x
andThreadIdx.x)
---
[Visit
Topic](https://discuss.tvm.apache.org/t/ops-become-slow-when-using-te-var/11486/4)
to respond.
You are receiving this because you en
Yes,I guess so.However for most case I think those ifs are unnecessary.So I
want to know the assertion sentences to avoid them.
Here's an example. `N` is the `te.var`.You can clearly see the duplicate `if`
.
extern "C" __global__ void default_function_kernel0(float* __restrict__
T_s
[quote="Maxwell-Hu, post:1, topic:4347, full:true"]
I'm interested in heterogeneous execution on multiple GPU, the basic idea is to
schedule different ops to different GPU.
What I expected was:
- GPU-0 executes 'sigmoid' and 'tanh'
- GPU-1 executes 'nn.dense'
However, the result seems that all
20 matches
Mail list logo
