[nnc] Fix special case of bias layer after fc layer in ACL backend (#2448)
authorEfimov Alexander/AI Tools Lab/./Samsung Electronics <a.efimov@samsung.com>
Fri, 30 Nov 2018 12:24:54 +0000 (15:24 +0300)
committerРоман Михайлович Русяев/AI Tools Lab /SRR/Staff Engineer/삼성전자 <r.rusyaev@samsung.com>
Fri, 30 Nov 2018 12:24:54 +0000 (15:24 +0300)
commit5bdc619d4b4c3f7f8abaf9ee44b374585963d78d
tree3dbcb7e81a76df2ddd34cb7c0ce759106f84ab22
parent37c732c749b23c4c2c9fbe6eac7e8ef8d2dae69c
[nnc] Fix special case of bias layer after fc layer in ACL backend (#2448)

Handle shape of bias weights with respect to previous operation.
Need to use different shape for operations after fully connected layer,
because it restores batch dimension in output tensor

Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
contrib/nnc/passes/acl_soft_backend/AclCppOpGenerator.cpp