arch-gcn3: Add memcpy condition when writing EXEC_LO

arch-gcn3: Add memcpy condition when writing EXEC_LO

Some compilers emit an error on the operand template class when writing
exec mask. Add a condition to explicitly set memcpy size argument to
32b or 64b based on the number of dwords.

Change-Id: I49b0e4a1680283e772d0a5a8efd687b31d4f1624
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29961
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>