Multi-core should significantly improve performance. Can you use htop to watch
the cpu usage?
---
[Visit
Topic](https://discuss.tvm.ai/t/tvm-is-10x-slower-than-simple-mxnet-model/7154/6)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these
How many CPU cores are you using? Latency number looks like single core
inference.
---
[Visit
Topic](https://discuss.tvm.ai/t/tvm-is-10x-slower-than-simple-mxnet-model/7154/4)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails,
Change target to "llvm -mcpu=skylake-avx512"?
---
[Visit
Topic](https://discuss.tvm.ai/t/tvm-is-10x-slower-than-simple-mxnet-model/7154/2)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
https://tvm.apache.org/docs/dev/relay_add_op.html
---
[Visit
Topic](https://discuss.tvm.ai/t/graph-tuning-a-te-computation-te-op-to-relay-expr/7037/5)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
@cu Would you mind file a patch for this?
---
[Visit
Topic](https://discuss.tvm.ai/t/class-tvm-relay-expr-call-has-no-attribute-name-hint-when-importing-ssd-resnet34-from-tensorflow/6989/6)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from
This means padlist is actually symbolic. You can try something like ```padlist
= _infer_value(inputs[1], params, mod)``` to see whether it can be pre-computed.
---
[Visit
Can you print full trace?
---
[Visit
Topic](https://discuss.tvm.ai/t/class-tvm-relay-expr-call-has-no-attribute-name-hint-when-importing-ssd-resnet34-from-tensorflow/6989/2)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails,
https://github.com/apache/incubator-tvm/issues/5710
---
[Visit
Topic](https://discuss.tvm.ai/t/bug-tvm-generate-wrong-opencl-code/6658/5) to
respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
Dynamic input shape will involve dynamic shape kernel codegen, which is still
WIP.
---
[Visit
Topic](https://discuss.tvm.ai/t/how-to-setting-model-compiled-from-pytorch-with-mutable-input-size/6827/4)
to respond.
You are receiving this because you enabled mailing list mode.
To
It looks like ```multiply``` is not recognized as mutiple input op. Can you dig
into
https://github.com/apache/incubator-tvm/blob/master/python/tvm/autotvm/graph_tuner/base_graph_tuner.py#L167
to see why it is not recognized?
---
[Visit
What is your target device?
---
[Visit
Topic](https://discuss.tvm.ai/t/question-after-auto-tune-the-speed-of-resnest101-and-resnet101-have-big-difference/6748/2)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
I'm refactoring tf frontend tensor array and will fix these issues.
---
[Visit Topic](https://discuss.tvm.ai/t/issue-with-static-tensor-array/6333/12)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
Integrating MKLDNN into deep learning framework requires special handle for the
fusion of those operators which will be accelerated using MKLDNN kernel,
sometimes even patten matching nn blocks. It's quite difficult to develop a
general fusion rule in this situation.
---
[Visit
Graph tuning is to balance the tradeoff between data layout overhead and kernel
with fast schedule, MKLDNN needs to be integrated with deep learning framework
to execute a NN. In this process, overheads such as data layout transformation
can generate. Another major overhead comes from
@moderato The performance advantage comes from the joint optimization of both
kernel and graph level optimization. For conv2d kernels, in lots of cases
MKLDNN outperforms TVM. However, with graph tuning we can achieve better e2e
performance by carefully arranging data layout.
---
[Visit
You don't need graph tuning while using cblas.
---
[Visit
Topic](https://discuss.tvm.ai/t/can-tvm-now-support-batched-inference-autotvm-runs-twice-as-long-as-tensorflow/6405/8)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails,
I'm fixing some issues regarding to tf ssd models and will submit a PR soon.
---
[Visit Topic](https://discuss.tvm.ai/t/issue-with-static-tensor-array/6333/6)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
Can you provide the full trace?
Also it should be related to certain tensor array op. It would be better if you
can provide a minimal case.
---
[Visit Topic](https://discuss.tvm.ai/t/issue-with-static-tensor-array/6333/2)
to respond.
You are receiving this because you enabled mailing
Currently log files in tophub just store the best schedule for each workload.
The idea for graph tuner is to select a schedule from topk(usually 20 - 30)
best schedules from a workload so that we can minimize data layout
transformation overhead. Thus we want to first do autotuning.
---
This is a quite valuable topic which can help us figure out what kind of
information related to optimization we can get from TVM IR itself, after all
the LLVM optimization pass applied. For x86 conv2d, my observation is that the
work llvm unrolling is doing can be implemented in TVM schedule
Did you call cfg.add_flop to add flops number?
---
[Visit
Topic](http://tracking.discuss.tvm.ai/tracking/click?d=2zHJ-oLuPzy-Sp0s1DTQwn9qeKjUgBbdu3W75xbqGy881gxEFiNQ5hRjAKkz5tGT9Ou4R8dj6zHjz4rG9XUgXoJYT0a7H5b0HfcHZnxJbQFbREDwZmkJ7PDOWLim8tqzaQFhKHtut8CeeAGHTonhFNJZLxxYTwUdiJl-J8aLaq9G0)
You can print the workload of each conv2d: tsk.workload.
---
[Visit
Topic](https://discuss.tvm.ai/t/how-to-map-lowered-function-with-task-name-in-autotvm/3704/2)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
22 matches
Mail list logo
