-
Notifications
You must be signed in to change notification settings - Fork 9
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
* added protobuf module (purely in C, based on protobuf-c) and Onnx p…
…arsing module based on it + simple demo script * extended compiler to simplify creation of complex modules that include big chunks of C/C++ code * fixed a few bugs in the tutorial
- Loading branch information
Showing
14 changed files
with
10,616 additions
and
13 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Binary file not shown.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,13 @@ | ||
| /* | ||
| This file is a part of ficus language project. | ||
| See ficus/LICENSE for the licensing terms | ||
| */ | ||
|
|
||
| // dumps .onnx model | ||
|
|
||
| import Onnx.Ast, Onnx.Parse, Sys | ||
| val model_name = match Sys.arguments() { | ||
| | f :: [] => f | ||
| | _ => println("specify .onnx model name"); throw Fail("") | ||
| } | ||
| println(Onnx.Parse.parse(model_name)) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,215 @@ | ||
| /* | ||
| This file is a part of ficus language project. | ||
| See ficus/LICENSE for the licensing terms | ||
| */ | ||
|
|
||
| // hierarchical representation of Onnx format | ||
| import Hashmap | ||
|
|
||
| type datatype_t = | ||
| | DTYP_UNDEFINED | DTYP_FLOAT | DTYP_UINT8 | DTYP_INT8 | DTYP_UINT16 | DTYP_INT16 | ||
| | DTYP_INT32 | DTYP_INT64 | DTYP_STRING | DTYP_BOOL | DTYP_FLOAT16 | DTYP_DOUBLE | ||
| | DTYP_UINT32 | DTYP_UINT64 | DTYP_COMPLEX64 | DTYP_COMPLEX128 | DTYP_BFLOAT16 | ||
|
|
||
| type tdata_t = T_FLOAT: float [] | T_INT8: int8 [] | T_INT64: int64 [] | ||
| type tensor_id_t = string | ||
|
|
||
| type tensor_t = | ||
| { | ||
| name: tensor_id_t | ||
| shape: int [] | ||
| data: tdata_t | ||
| } | ||
|
|
||
| type opset_t = | ||
| { | ||
| version: int64 | ||
| domain: string | ||
| } | ||
|
|
||
| type dim_t = DimParam: string | DimValue: int64 | ||
| type typeinfo_t = TYPINFO_TENSOR: (datatype_t, dim_t []) | ||
|
|
||
| type valueinfo_t = | ||
| { | ||
| name: tensor_id_t | ||
| denotation: string | ||
| typ: typeinfo_t | ||
| } | ||
|
|
||
| type attrval_t = | ||
| | AttrInt: int64 | AttrFloat: float | ||
| | AttrString: string | AttrTensor: tensor_t | ||
| | AttrFloats: float [] | AttrInts: int [] | AttrStrings: string [] | ||
|
|
||
| type attr_t = | ||
| { | ||
| name: string | ||
| v: attrval_t | ||
| } | ||
|
|
||
| type node_t = | ||
| { | ||
| name: string | ||
| op: string | ||
| inputs: tensor_id_t [] | ||
| outputs: tensor_id_t [] | ||
| attrs: attr_t [] | ||
| } | ||
|
|
||
| type graph_t = | ||
| { | ||
| name: string | ||
| inputs: valueinfo_t [] | ||
| outputs: valueinfo_t [] | ||
| values: valueinfo_t [] | ||
| initializers: tensor_t [] | ||
| nodes: node_t [] | ||
| } | ||
|
|
||
| type model_t = | ||
| { | ||
| ir_version: int64 | ||
| producer: string | ||
| domain: string | ||
| doc_string: string | ||
| import_opsets: opset_t [] | ||
| metadata: (string, string) [] | ||
| graph: graph_t | ||
| } | ||
|
|
||
| fun total(t: tensor_t) = product(t.shape, 1) | ||
|
|
||
| fun string(d: datatype_t) | ||
| { | ||
| | DTYP_UNDEFINED => "undefined" | ||
| | DTYP_FLOAT => "float" | ||
| | DTYP_UINT8 => "uint8" | ||
| | DTYP_INT8 => "int8" | ||
| | DTYP_UINT16 => "uint16" | ||
| | DTYP_INT16 => "int16" | ||
| | DTYP_INT32 => "int32" | ||
| | DTYP_INT64 => "int64" | ||
| | DTYP_STRING => "string" | ||
| | DTYP_BOOL => "bool" | ||
| | DTYP_FLOAT16 => "float16" | ||
| | DTYP_DOUBLE => "double" | ||
| | DTYP_UINT32 => "uint32" | ||
| | DTYP_UINT64 => "uint64" | ||
| | DTYP_COMPLEX64 => "complex64" | ||
| | DTYP_COMPLEX128 => "complex128" | ||
| | DTYP_BFLOAT16 => "bfloat16" | ||
| } | ||
|
|
||
| fun string(ops: opset_t) = f"{ops.domain} v{ops.version}" | ||
| fun string(dim: dim_t) { | ||
| | DimParam(n) => n | ||
| | DimValue(v) => string(v) | ||
| } | ||
|
|
||
| fun tensor_data_prefix(t: tdata_t) | ||
| { | ||
| | T_FLOAT _ => "float32" | ||
| | T_INT8 _ => "int8" | ||
| | T_INT64 _ => "int64" | ||
| } | ||
|
|
||
| fun print_tensor_data(t: tdata_t) | ||
| { | ||
| | T_FLOAT(data) => print(data) | ||
| | T_INT8(data) => print(data) | ||
| | T_INT64(data) => print(data) | ||
| } | ||
|
|
||
| fun print(t: tensor_t) | ||
| { | ||
| val shape = "x".join(t.shape.map(string)) | ||
| print(f"tensor {{name='{t.name}', shape={shape}, data={tensor_data_prefix(t.data)} ") | ||
| if total(t) > 10 {print("[...]")} else {print_tensor_data(t.data)} | ||
| print("}") | ||
| } | ||
|
|
||
| fun print(a: attr_t) | ||
| { | ||
| print(f"{a.name}: ") | ||
| match a.v { | ||
| | AttrInt(i) => print(i) | ||
| | AttrFloat(f) => print(f) | ||
| | AttrString(s) => print(repr(s)) | ||
| | AttrTensor(t) => print(t) | ||
| | AttrInts(ints) => print(ints) | ||
| | AttrFloats(floats) => print(floats) | ||
| | AttrStrings(strings) => print(strings) | ||
| } | ||
| } | ||
|
|
||
| fun print(n: node_t) | ||
| { | ||
| val indent0 = " "*4 | ||
| val indent1 = indent0 + indent0 | ||
| val indent2 = indent1 + indent0 | ||
| val indent3 = indent2 + indent0 | ||
| println(f"{{") | ||
| println(f"{indent2}name: '{n.name}'") | ||
| println(f"{indent2}op: {n.op}") | ||
| println(f"{indent2}inputs: {n.inputs}") | ||
| println(f"{indent2}outputs: {n.outputs}") | ||
| print(f"{indent2}attributes: {{") | ||
| if n.attrs.empty() {print(f"}}\n{indent1}}}")} | ||
| else { | ||
| println() | ||
| for a <- n.attrs { | ||
| print(indent3) | ||
| println(a) | ||
| } | ||
| print(f"{indent2}}}\n{indent1}}}") | ||
| } | ||
| } | ||
|
|
||
| fun print(vi: valueinfo_t) | ||
| { | ||
| print(f"{vi.name}") | ||
| if vi.denotation != "" { | ||
| print(f" ({vi.denotation})") | ||
| } | ||
| print(": ") | ||
| match vi.typ { | ||
| | TYPINFO_TENSOR (dt, diminfo) => | ||
| val diminfo = " x ".join(diminfo.map(string)) | ||
| print(f"{dt}, {diminfo}") | ||
| } | ||
| } | ||
|
|
||
| fun print(graph: graph_t) | ||
| { | ||
| val indent0 = " "*4, indent1 = indent0 + indent0 | ||
| println(f"graph {{") | ||
| println(f"{indent0}name: {graph.name},") | ||
| println(f"{indent0}inputs: {{") | ||
| for x <- graph.inputs { print(indent1); print(x); println(",") } | ||
| println(f"{indent0}}},\n{indent0}outputs: {{") | ||
| for x <- graph.outputs { print(indent1); print(x); println(",") } | ||
| println(f"{indent0}}},\n{indent0}values: {{") | ||
| for x <- graph.values { print(indent1); print(x); println(",") } | ||
| println(f"{indent0}}},\n{indent0}initializers: {{") | ||
| for x <- graph.initializers { print(indent1); print(x); println(",") } | ||
| println(f"{indent0}}},\n{indent0}nodes: {{") | ||
| for x <- graph.nodes { print(indent1); print(x); println(",") } | ||
| print(f"{indent0}}}\n}") | ||
| } | ||
|
|
||
| fun print(model: model_t) | ||
| { | ||
| println(f"ir_version: {model.ir_version}") | ||
| println(f"producer: {model.producer}") | ||
| println(f"domain: {model.domain}") | ||
| println(f"doc_string: '{model.doc_string}'") | ||
| for opset <- model.import_opsets { | ||
| println(f"import opset: {opset}") | ||
| } | ||
| for (k, v) <- model.metadata { | ||
| println(f"property '{k}': '{v}'") | ||
| } | ||
| println(model.graph) | ||
| println("// end of model") | ||
| } |
Oops, something went wrong.