[TF FE]: Support complex tensors for Equal operation #29339
Conversation
github-actions
bot
added
category: Core
| @@ -289,6 +291,52 @@ ov::Output<ov::Node> ComplexTypeMark::div(const NodeContext& context, | |||
| return {result}; | |||
| } | |||
|
|
|||
| ov::Output<ov::Node> ComplexTypeMark::equal(const NodeContext& context, | |||
There was a problem hiding this comment.
please move this functionality to common translators. We will re-use it in PyTorch FE.
There was a problem hiding this comment.
Should I move it to src\frontends\common_translators\src\op\complex.cpp or put it in a new file?
There was a problem hiding this comment.
new file equal.cpp in src\frontends\common_translators\src\op\
| auto lr = lhs_complex->get_real(); | ||
| auto li = lhs_complex->get_imag(); | ||
| auto rr = rhs_complex->get_real(); | ||
| auto ri = rhs_complex->get_imag(); |
There was a problem hiding this comment.
please think how to implement it using get_data() and how it will simplify this decomposition
| } else if (lhs_complex) { | ||
| // rhs is of a real type | ||
| auto lhs_real = lhs_complex->get_real(); | ||
| auto lhs_imag = lhs_complex->get_imag(); | ||
|
|
||
| auto eq_real = context.mark_node(make_shared<v1::Equal>(lhs_real, rhs)); | ||
| auto zero_const = context.mark_node(make_shared<v0::Constant>(lhs_imag.get_element_type(), Shape{}, 0)); | ||
| auto eq_imag = context.mark_node(make_shared<v1::Equal>(lhs_imag, zero_const)); | ||
|
|
||
| auto result = context.mark_node(make_shared<v1::LogicalAnd>(eq_real, eq_imag)); | ||
| return {result}; | ||
| } else if (rhs_complex) { | ||
| // lhs is of a real type | ||
| auto rhs_real = rhs_complex->get_real(); | ||
| auto rhs_imag = rhs_complex->get_imag(); | ||
|
|
||
| auto eq_real = context.mark_node(make_shared<v1::Equal>(lhs, rhs_real)); | ||
| auto zero_const = context.mark_node(make_shared<v0::Constant>(rhs_imag.get_element_type(), Shape{}, 0)); | ||
| auto eq_imag = context.mark_node(make_shared<v1::Equal>(rhs_imag, zero_const)); | ||
|
|
||
| auto result = context.mark_node(make_shared<v1::LogicalAnd>(eq_real, eq_imag)); | ||
| return {result}; | ||
| } |
There was a problem hiding this comment.
these two cases are not supported by TF. Both operands must be of the same type: https://www.tensorflow.org/api_docs/python/tf/raw_ops/Equal
github-actions
bot
added
no-match-files
and removed
category: CPP API
| @@ -8,6 +8,7 @@ | |||
| #include <map> | |||
| #include <string> | |||
|
|
|||
| #include "common_translators.hpp" | |||
There was a problem hiding this comment.
not needed
| #include "common_translators.hpp" |
| @@ -19,6 +19,8 @@ COMMON_OP_CONVERTER(translate_imag); | |||
| COMMON_OP_CONVERTER(translate_atan2); | |||
| COMMON_OP_CONVERTER(translate_angle); | |||
|
|
|||
| COMMON_OP_CONVERTER(translate_equal_op); | |||
There was a problem hiding this comment.
| COMMON_OP_CONVERTER(translate_equal_op); | |
| COMMON_OP_CONVERTER(translate_equal); |
let us have names without _op suffix. It is obvious that this is for ops:)
|
@mvafin, fyi. it can be reused for PT FE |
|
build_jenkins |
| OutputVector translate_equal_op(const NodeContext& node) { | ||
| default_op_checks(node, 2, {"Equal"}, true); | ||
|
|
||
| auto result = common_translators::translate_equal(node); |
There was a problem hiding this comment.
@rkazants It seems that remove #include "common_translators.hpp" from src/frontends/tensorflow_common/include/common_op_table.hpp will result in an error here?
There was a problem hiding this comment.
Any suggestions about the implementation of translate_equal_op function here?
There was a problem hiding this comment.
please include "common_translators.hpp" in "binary_op.cpp". No need to insert to header file, we use it internally in this case.
|
@rkazants Updated |
|
build_jenkins |
Details:
Tickets: