MLBEDSW-11040 Fix ARGMAX decompose
Fixed ARGMAX decompose of large tensor dimensions caused by slice information lost for chained ADD operations. Also reduced block size for ARGMAX decompose to avoid possible negative indices, due to the way the output is shifted.
Change-Id: I21f8309343fe6019a49d0c760711d28acdfb5b13 Signed-off-by: Fredrik Svedberg fredrik.svedberg@arm.com