Skip to content

N
ngraph
Unverified Commit 5a656a1a authored by Adam Osewski's avatar Adam Osewski Committed by GitHub
Browse files
Options

[ONNX] Update status of supported ops. (#4292) 

* Update table.

* More updates.
Co-authored-by: 's avatarMichał Karzyński <postrational@users.noreply.github.com>
Co-authored-by: 's avatarScott Cyphers <diyessi@users.noreply.github.com>
Co-authored-by: 's avatarRobert Kimball <robert.kimball@intel.com>
parent 2b715228
Show whitespace changes
Inline Side-by-side
Showing with 152 additions and 109 deletions
src/ngraph/frontend/onnx_import/op/supported_ops.md
View file @ 5a656a1a
Generally nGraph doesn't support tensors of types: Generally nGraph doesn't support tensors of types:
* `string`, * `string`,
* `float16`,
* `complex64`, * `complex64`,
* `complex128`. * `complex128`.
...@@ -16,107 +15,140 @@ opset versions starting from `1` to `6` and to the latest opset version. ...@@ -16,107 +15,140 @@ opset versions starting from `1` to `6` and to the latest opset version.
## Supported Ops: ## Supported Ops:
| Name | Opset supported | Comment | | Name | ONNX Opset supported | nGraph opset support | Comment |
|------|----------------------------|---------| |------|----------------------------|---------|-----|
| Abs | 1-6- | | Abs | 1-6- | 0,1 |
| Acos | 7- | | Acos | 7- | 0,1 |
| Acosh | 9- | | Acosh | 9- | 0, | Have to change to only v1 ops (NGONNX-1015)
| Add | (1-6-)7- | Legacy broadcasting rules not supported. (NGONNX-496) | Add | 1-6-7- | 0,1 |
| And | 1-7- | | And | 1-7- | 0,1 |
| ArgMax | 1- | | ArgMax | 1- | 0,1 |
| ArgMin | 1- | | ArgMin | 1- | 0,1 |
| Asin | 7- | | Asin | 7- | 0,1 |
| Asinh | 9- | | Asinh | 9- | 0, | Have to change to only v1 ops (NGONNX-1015)
| Atan | 7 - | | Atan | 7 - | 0,1 |
| Atanh | 9- | | Atanh | 9- | 0, | Have to change to only v1 ops (NGONNX-1015)
| AveragePool | 1-7- | | AveragePool | 1-7- | 0,1 |
| BatchNormalization | 1-6-7- | | BatchNormalization | 1-6-7- | 0,1 |
| Ceil | 1-6- | | Cast | 1-6-9- | 0,1 |
| Clip | 1-6- | | Ceil | 1-6- | 0,1 |
| Concat | 1-4- | | Clip | 1-6- | 0,1 |
| Constant | 1- | | Concat | 1-4- | 0,1 |
| Conv | 1- | | Constant | 1- | 0,1 |
| ConvTranspose | 1- | | Conv | 1- | 0,1 |
| Cos | 7- | | ConvInteger | 10- | 0, |
| Cosh | 9- | | ConvTranspose | 1- | 0,1 |
| Div | (1-6-)7- | Legacy broadcasting rules not supported. (NGONNX-496) | Cos | 7- | 0,1 |
| Dropout | 1-6-7- | Only for inference. | Cosh | 9- | 0,1 | Have to change to only v1 ops (NGONNX-1015)
| Elu | 1-6- | | CumSum | 11- | 0, | NGONNX-944
| Equal | 1-7 | | DepthToSpace | 1-11- | 0,1 |
| Exp | 1-6- | | DequantizeLinear | 10- | 0, |
| Flatten | 1-9- | | Div | 1-6-7- | 0,1 |
| Floor | 1-6- | | Dropout | (1-6-7)-10- | 0,1 | Only for inference.
| Gemm | 1-6-7-9 | Some tests failing (NGONNX-494) | Elu | 1-6- | 0,1 |
| GlobalAveragePool | 1- | | Equal | 1-7 | 0,1 |
| GlobalLpPool | 1-2- | | Erf | 9- | 0,1 |
| GlobalMaxPool | 1- | | Exp | 1-6- | 0,1 |
| Greater | 1-7-9 | | Expand | 8- | 0,1 | Only static version
| HardSigmoid | 1-6- | | EyeLike | 9- | 0,1 |
| Identity | 1- | | Flatten | 1-9- | 0,1 |
| InstanceNormalization | 1- | | Floor | 1-6- | 0,1 |
| LRN | 1- | | Gather | 11- | 0,1 |
| LeakyRelu | 1-6- | | GatherND | 11- | 0, |
| Less | 1-7-9 | | Gemm | 1-6-7-9-11 | 0, | Some tests failing (NGONNX-494), Have to change to only v1 ops (NGONNX-1015)
| Log | 1-6- | | GlobalAveragePool | 1- | 0,1 |
| LogSoftmax | 1- | | GlobalLpPool | 1-2- | 0, (1) | Not fully v1, need `lp_norm` expressed with v1 ops (NGONNX-1018)
| LpNormalization | 1- | | GlobalMaxPool | 1- | 0,1 |
| MatMul | 1-9 | | Greater | 1-7-9 | 0,1 |
| Max | 1-6-8- | | HardSigmoid | 1-6- | 0,1 |
| MaxPool | 1-8- | | Hardmax | 11- | 0, (1) | GOE
| Mean | 1-6-8- | | Identity | 1- | 0,1 |
| Min | 1-6-8- | | InstanceNormalization | 1- | 0, (1) | Have to change to only v1 ops (NGONNX-1015)
| Mul | (1-6-)7- | Legacy broadcasting rules not supported. (NGONNX-496) | LRN | 1- | 0,1 |
| Neg | 1-6- | | LSTM | 1-7- | 0,(1) | (NGONNX-1015), there is `GetOutputElement` (3 outputs), LSTMSequence and LSTMCell uses some v0 ops which aren't used in v1
| Not | 1- | | LeakyRelu | 1-6- | 0,(1) | (NGONNX-1015)
| OneHot | (9) | Only static version | Less | 1-7-9 | 0,1 |
| Or | 1-7- | | Log | 1-6- | 0,1 |
| PRelu | 1-6-7-9 | | LogSoftmax | 1- | 0,1 |
| Pow | 1-7- | | LpNormalization | 1- | 0, | (NGONNX-1018) need to update some builders
| Reciprocal | 1-6- | | MatMul | 1-9 | 0,(1) | Uses `v0::Dot`, v0 broadcasts and reshapes, update builders
| ReduceL1 | 1- | | MatMulInteger | 10- | 0, | `v0::QuantizedDot`
| ReduceL2 | 1- | | Max | 1-6-8- | 0, 1 |
| ReduceLogSum | 1- | | MaxPool | 1-8- | 0, 1 |
| ReduceLogSumExp | 1- | | Mean | 1-6-8- | 0, 1 |
| ReduceMax | 1- | | Min | 1-6-8- | 0,1 |
| ReduceMean | 1- | | Mod | 10- | 1 |
| ReduceMin | 1- | | Mul | (1-6-)7- | 0,1 |
| ReduceProd | 1- | | Neg | 1-6- | 0,1 |
| ReduceSum | 1- | | NonMaxSuppression | 11- | 1 |
| ReduceSumSquare | 1- | | Not | 1- | 0,1 | (aka `v1::LogicalNot`)
| Relu | 1-6- | | OneHot | (9) | 0, (1) | (NGONNX-1015)
| Selu | 1-6- | | Or | 1-7- | 0,1 | (aka `v1::LogicalOr`)
| Shape | 1- | | PRelu | 1-6-7-9 | 0, 1 | fused op uses arithmetic and broadcasting from v0
| Shrink | 1- | | Pad | 1-2-11- | 0, (1) | (NGONNX-1015)
| Sigmoid | 1-6- | | Pow | 1-7- | 0,1 |
| Sign | 9- | | QLinearConv | 10- | 0 | `opset0::QuantizedConvolution`
| Sin | 7- | | QLinearMatMul | 10- | 0 | `v0::QuantizedDot`
| Sinh | 9- | | QuantizeLinear | 10- | 0 | `opset0::Quantize`
| Size | 1- | | Reciprocal | 1-6- | 0, 1|
| Slice | 1- | | ReduceL1 | 1- | 0, | (NGONNX-1018)
| Softmax | 1- | | ReduceL2 | 1- | 0,1 |
| Softplus | 1- | | ReduceLogSum | 1- | 0,1 |
| Softsign | 1- | | ReduceLogSumExp | 1- | 0,1 |
| Split | 1-2- | | ReduceMax | 1- | 0,1 |
| Sqrt | 1-6- | | ReduceMean | 1- | 0,1 |
| Squeeze | 1- | | ReduceMin | 1- | 0,1 |
| Sub | (1-6-)7- | Legacy broadcasting rules not supported. (NGONNX-496) | ReduceProd | 1- | 0,1 |
| Sum | 1-6-8- | | ReduceSum | 1- | 0,1 |
| Tan | 7- | | ReduceSumSquare | 1- | 0,1 |
| Tanh | 1-6- | | Relu | 1-6- | 0,1 |
| TopK | 1- | | Reshape | 1-5- | (0,1) | v1 supports dynamic target shape, but only as Constant?
| Transpose | 1- | | ReverseSequence | 10- | 0,1 |
| Unsqueeze | 1- | | ScatterND | 11- | 0, |
| Xor | 1-7- | | Selu | 1-6- | 0, 1 |
| Where | 9- | | Shape | 1- | 0,1 |
| Shrink | 1- | 0,1 |
| Sigmoid | 1-6- | 0,1 |
| Sign | 9- | 0,1 |
| Sin | 7- | 0,1 |
| Sinh | 9- | 0,1 |
| Size | 1- | 0,1 |
| Slice | 1- | 0,1 |
| Softmax | 1- | 0,1 |
| Softplus | 1- | 0,1 |
| Softsign | 1- | 0,(1) | (NGONNX-1015)
| SpaceToDepth | 1- | 0,1 |
| Split | 1-2- | 0,1 |
| Sqrt | 1-6- | 0,1 |
| Squeeze | 1- | 0,(1) |
| Sub | (1-6-)7- | 0,1 |
| Sum | 1-6-8- | 0,1 |
| Tan | 7- | 0,1 |
| Tanh | 1-6- | 0,1 |
| ThresholdedRelu | 10- | 0,1 |
| TopK | 1- | 0,(1) | Need v0::GOE
| Transpose | 1- | 0,1 |
| Unsqueeze | 1- | 0,1 |
| Xor | 1-7- | 0,1 |
| Where | 9- | 0,1 |
### Able to implement or WIP
| Name | Opset supported | NGCORE | NGONNX | Comment |
|------|-----------------|--------|--------|---------|
| BitShift | (11)- | | 1014 |
| ConstantOfShape | (9) | 286 | 445 | Dynamic shape input. WIP |
| DynamicQuantizeLinear | (11) | | 786 |
| GRU | - | | 325, 177 | There is no `GRUCell` nor `GRU` in v1 |
| RNN | - | | 323, 287 | `v1::RNNCell`|
| Round | (11)- | | 1008 | `v0::Round`
| Tile | - | NGRAPH-3292 | 368 | Dynamic op. WIP |
| Cast | 1-6- | 290 | 452 | Float16 unsupported. |
## Unsupported Ops: ## Unsupported Ops:
### Lack of support in nGraph ### Lack of support in nGraph
| Name | Opset supported | NGCORE | NGONNX | Comment | | Name | Opset supported | NGCORE | NGONNX | Comment |
|------|-----------------|--------|--------|---------| |------|-----------------|--------|--------|---------|
| Erf | (9) | 284 | 489 | Need separate kernel for this in nGraph core. |
| Pad | 1-2- | 273 | 416, 498 | Not fully supported. |
| LSTM | 1-7- | | 476 | Mixed sequences length not supported yet. |
| MaxUnpool | (9) | 286, 289 | 447 | | | MaxUnpool | (9) | 286, 289 | 447 | |
| LpPool | - | 291 | 488 | Unsupported by nGraph - only max/avg pooling ops. Need separate kernel. | | LpPool | - | 291 | 488 | Unsupported by nGraph - only max/avg pooling ops. Need separate kernel. |
| Multinomial | - | 199 | 435 | Lack of PRNG in nGraph. | | Multinomial | - | 199 | 435 | Lack of PRNG in nGraph. |
...@@ -124,35 +156,46 @@ opset versions starting from `1` to `6` and to the latest opset version. ...@@ -124,35 +156,46 @@ opset versions starting from `1` to `6` and to the latest opset version.
| RandomNormalLike | - | 199 | 434 | Lack of PRNG in nGraph. | | RandomNormalLike | - | 199 | 434 | Lack of PRNG in nGraph. |
| RandomUniform | - | 199 | 434 | Lack of PRNG in nGraph. | | RandomUniform | - | 199 | 434 | Lack of PRNG in nGraph. |
| RandomUniformLike | - | 199 | 434 | Lack of PRNG in nGraph. | | RandomUniformLike | - | 199 | 434 | Lack of PRNG in nGraph. |
| Cast | 1-6- | 290 | 452 | Float16 unsupported. | | IsInf | (10) | | 528 |
| StringNormalizer | (10) | | 600 | Need support for `string` data type.
| TfIdfVectorizer | (9) | | 523 |
| Det | (11) | | 754 | |
### Futher analysis needed ### Futher analysis needed
| Name | Opset supported | NGCORE | NGONNX | Comment | | Name | Opset supported | NGCORE | NGONNX | Comment |
|------|-----------------|--------|--------|---------| |------|-----------------|--------|--------|---------|
| GRU | - | | 325, 177 | Should be possible to implement. Look at `LSTM` |
| RNN | - | | 323, 287 | Should be similar to `LSTM`. |
| If | - | | 432 | At this moment probably impossible. | | If | - | | 432 | At this moment probably impossible. |
| IsNaN | (9) | | 440 | Hacky way is to generate constant nodes with representations of NaN and compare with them. | | IsNaN | (9) | | 440 | Hacky way is to generate constant nodes with representations of NaN and compare with them. |
| Loop | - | | 432 | Static loops with some preconditions may be possible, however no idea how to pass graph (proto?) as a _body_ attribute. (what about graph contains `Loop`?) | | Loop | - | | 432 | Static loops with some preconditions may be possible, however no idea how to pass graph (proto?) as a _body_ attribute. (what about graph contains `Loop`?) |
| Scan | - | | 433 | Further analysis needed. - determine whether it is possible to import graph passed by op attribute. | | Scan | - | | 433 | Further analysis needed. - determine whether it is possible to import graph passed by op attribute. |
| Einsum | (12) | | | User can define in a language the operation to perform |
| NonZero | (9) | | 472 | Maybe we can leverage TopK here? First count NonZero elements with logic ops and reduction and then TopK.
| Resize | (10-11)- | | 782 | Look like Interpolation over ROIs. Very specialized types of interpolation.
| ScatterElements | (11) | | 977 |
| ScatterND | (11) | | 1020 |
| Unique | (11) | | 761 |
### Dynamic operators ### Dynamic operators
| Name | Opset supported | NGCORE | NGONNX | Comment | | Name | Opset supported | NGCORE | NGONNX | Comment |
|------|-----------------|--------|--------|---------| |------|-----------------|--------|--------|---------|
| Compress | (9) | 285 | 438. | Dynamically selected indices | | Compress | (9-11) | 285 | 438 | Dynamically selected indices |
| ConstantOfShape | (9) | 286 | 445 | Dynamic shape input. |
| Expand | - | NGRAPH-3289 | 367 | Dynamic op. | | Expand | - | NGRAPH-3289 | 367 | Dynamic op. |
| Gather | - | NGRAPH-3291 | 369, | Dynamic op. | | GatherElements | - | | 757 | |
| OneHot | (9) | NGCORE-339 | 486 | Dynamic output shape | OneHot | (9) | NGCORE-339 | 486 | Dynamic output shape
| Tile | - | NGRAPH-3292 | 368 | Dynamic op. | | Upsample | (7-9-10-) | 287 | 441 | Dynamic op. **Deprecated** from opset 10 |
| Upsample | (7) | 287 | 441 | Dynamic op. |
| MaxRoiPool | - | 288 | 487 | Dynamic op - Need dynamic slicing. Beside just use _slice/op/concat_ pattern. | | MaxRoiPool | - | 288 | 487 | Dynamic op - Need dynamic slicing. Beside just use _slice/op/concat_ pattern. |
| Reshape | 1-5- | NGRAPH-3290 | 357 | Lack of support for dynamic shape input. Only as a Constant or as an Initializer. | | Reshape | 1-5- | NGRAPH-3290 | 357 | Lack of support for dynamic shape input. Only as a Constant or as an Initializer. |
| Scatter | (9) | 289 | 446 | Dynamic indices input. | | Scatter | (9) | 289 | 446 | Dynamic indices input. **Deprecated** in ONNX standard |
| RoiAlign | (10) | | 601 |
### Able to implement or WIP ### Sequence* ops
| Name | Opset supported | NGCORE | NGONNX | Comment | | Name | Opset supported | NGCORE | NGONNX | Comment |
|------|-----------------|--------|--------|---------| |------|-----------------|--------|--------|---------|
| Add, Sub, Mul, Div | 1-6 | | | We currently don't support legacy broadcasting rules for binary ops. | | ConcatFromSequence | (11)- | | 1016 |
| Cast | 1-6- | | 427 | Errors while casting to bool | | SequenceAt | (11) | | 1021 | need further analysis |
| Hardmax | - | | 431 | Use make constant and Argmax. See `test_ops_unary.py::test_hardmax()` | | SequenceConstruct | (11) | | 1021 | need further analysis |
| SequenceEmpty | (11) | | 1021 | need further analysis |
| SequenceErase | (11) | | 1021 | need further analysis |
| SequenceInsert | (11) | | 1021 | need further analysis |
| SequenceLength | (11) | | 1021 | need further analysis |
| SplitToSequence | (11) | | 1021 | need further analysis |
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment