[Apache TVM Discuss] [Development] Pass sparse tensor to tvm.build
Another work around is to get rid off the sparse placeholder completely.
Instead use three standard Tensors for each element in the sparse tensor (i.e.
data, indices and index pointers). Then feed those (plus the X matrix) to
`topi.nn.sparse_dense` and things seem to work.
There's a working implementation in [this
repo](https://github.com/ceruleangu/Block-Sparse-Benchmark):
```
@autotvm.template("benchmark/block_sparse")
def block_sparse_template(W_sp_np_data_shape, W_sp_np_indices_shape,
W_sp_np_indptr_shape, X_np_shape):
W_data = te.placeholder(shape=W_sp_np_data_shape, dtype='float32',
name='W_data')
W_indices = te.placeholder(shape=W_sp_np_indices_shape, dtype='int32',
name='W_indices')
W_indptr = te.placeholder(shape=W_sp_np_indptr_shape, dtype='int32',
name='W_indptr')
X = te.placeholder(shape=X_np_shape, dtype='float32', name='X')
Y = topi.nn.sparse_dense(X, W_data, W_indices, W_indptr)
...
```
---
[Visit
Topic](https://discuss.tvm.apache.org/t/pass-sparse-tensor-to-tvm-build/7739/3)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
here](https://discuss.tvm.apache.org/email/unsubscribe/25c4d871ca6183c7b3dfc00b30bc3e7af3fd1797038b3df3e5e05de0977a7fd9).
[Apache TVM Discuss] [Development] Pass sparse tensor to tvm.build
Useful resource, thanks. I ended up fixing the approach in my 2nd example (using three `ndarray`s. Basically one can only pass sparse tensors that have the same sparsity pattern, not just the same level of sparsity as the placeholders you pass. Thus, when constructing the placeholder tensors for compilation, one needs to give them sizes from the sparse data you will pass to the compiled function. E.g. if your sparse data is in a SciPy CSR object called `W_sp_np`, your TVM placeholders would be constructed with: ``` W_data = te.placeholder(shape=W_sp_np.data.shape, dtype=str(W_sp_np.data.dtype), name='W_data') W_indices = te.placeholder(shape=W_sp_np.indices.shape, dtype=str(W_sp_np.indices.dtype), name='W_indices') W_indptr = te.placeholder(shape=W_sp_np.indptr.shape, dtype=str(W_sp_np.indptr.dtype), name='W_indptr') ``` My (unoptimised) sparse NCHW GEMM convolution is now working, I'll see if I can draft a tutorial about what I've learned once I've completed some other things. --- [Visit Topic](https://discuss.tvm.apache.org/t/pass-sparse-tensor-to-tvm-build/7739/4) to respond. You are receiving this because you enabled mailing list mode. To unsubscribe from these emails, [click here](https://discuss.tvm.apache.org/email/unsubscribe/3e53da98e0c95e548259aeda3ffd4bd5850ce63a0cc2019be6305eaa7a4d71ce).
[Apache TVM Discuss] [Development] Supporting CumSum from ONNX - Use te Scan op or develop from scratch?
Hi, I've just came across a model that requires support for ONNX CumSum op https://github.com/onnx/onnx/blob/master/docs/Operators.md#CumSum. The model comes from DETR object detection model https://github.com/facebookresearch/detr. Since this model doesn't need ad hoc object detection ops that are painful to support, I think it is a great fit for TVM. Our ONNX frontend (also PyTorch) only needs to implement Cumsum op. Since TVM has support for scan operation https://tvm.apache.org/docs/tutorials/language/scan.html#sphx-glr-tutorials-language-scan-py, I'm wondering if it is a good idea to implement Relay cumsum op on top of te scan, or implement a new topi operator from scratch. I also want to utilize scan primitive from thrust to support fast cumsum on cuda. @tqchen @kevinthesun @Laurawly @jwfromm --- [Visit Topic](https://discuss.tvm.apache.org/t/supporting-cumsum-from-onnx-use-te-scan-op-or-develop-from-scratch/7830/1) to respond. You are receiving this because you enabled mailing list mode. To unsubscribe from these emails, [click here](https://discuss.tvm.apache.org/email/unsubscribe/b1f0ea64da5c455627578bc564be6dcea38fc140d3c5c8902e08e38aa6daa68c).
