operator | versions | IsSupport | front-end support | IR support | Based Data type supported (float64/32/16, int64/int32/16) | unit-test passed? | comments
-- | -- | -- | -- | -- | -- | -- | --
Abs | 13, 6, 1 | Y | All versions | Y | all nnfusion's datatype | N(failed at unsupprted datatype uint8_t) | opset 13 update only extends bfloat16 support, nnfusion doesn't have this type
Acos | 7 | Y | All versions | Y | all nnfusion's datatype | Y |
Add | 14, 13, 7, 6, 1 | Y | All versions | Y | all nnfusion's datatype | Y | only datatype is different
And | 7, 1 | Y | All versions | Y | all nnfusion's datatype | Y |
ArgMax | 13, 12, 11, 1 | Y | All versions | Y | | N (but validated under hand customized code) | datatype issue; still not implemented keep_dims
ArgMin | 13, 12, 11, 1 | Y | All versions | Y | | N (but validated under hand customized code) | datatype issue; still not implemented keep_dims
Asin | 7 | Y | All versions | Y | all nnfusion's datatype | Y |
Atan | 7 | Y | All versions | Y | all nnfusion's datatype | Y |
test failed @ test_add_unit8 because currently do not support uint8_t datatype and it‘s not in the based datatype yet.
test of argmax and argmin still failed because of the datatype mismatch we found in this issue, but I customized the generated cuda project and compared with the output of numpy.argmax, numpy.argmin, and these two have the same behavior.
Unit Test Status
test failed @ test_add_unit8 because currently do not support uint8_t datatype and it‘s not in the based datatype yet.
test of argmax and argmin still failed because of the datatype mismatch we found in this issue, but I customized the generated cuda project and compared with the output of numpy.argmax, numpy.argmin, and these two have the same behavior.