arc: Do not pass -mcode-density to the assembler with -mno-code-density

[luismgsilva]

arc: Do not pass -mcode-density to the assembler with -mno-code-density

For ARC EM CPUs that enable code density by default, cpu_to_as adds
-mcode-density to the assembler command line.  When the user passes
-mno-code-density, the compiler disables the extension, but cpu_to_as
still requested -mcode-density for those CPUs.  The GNU assembler only
has a positive -mcode-density option and no way to disable the
extension from the command line, so the driver must omit -mcode-density
instead of forwarding the user's -mno-code-density flag.

Pass a marker from ASM_SPEC when -mno-code-density is given, and
suppress the default -mcode-density request in cpu_to_as.  Explicit
-mcode-density continues to be forwarded by %{mcode-density}.

For example, with -mcpu=em4_dmips the driver invokes the assembler as:

        gcc ... -mcpu=em4_dmips -c test.c
        -> as ... -mcpu=arcem -mcode-density ...

With -mno-code-density, it used to make no difference:

        gcc ... -mcpu=em4_dmips -mno-code-density -c test.c
        -> as ... -mcpu=arcem -mcode-density ...

and now correctly omits -mcode-density:

        gcc ... -mcpu=em4_dmips -mno-code-density -c test.c
        -> as ... -mcpu=arcem ...

Assembling the same test program shows different instruction encodings
in the object file, for example:

        $ gcc -c test.c -mcpu=em4_dmips
        $ objdump -d test.o
           4:   4104                    ld_s    r1,[r0,0]
          10:   4004                    ld_s    r0,[r0,0]

versus:

        $ gcc -c test.c -mcpu=em4_dmips -mno-code-density
        $ objdump -d test.o
           4:   8020                    ld_s    r1,[r0,0]
          10:   8000                    ld_s    r0,[r0,0]

gcc/ChangeLog:

        * config/arc/arc.h (ASM_SPEC): Pass a no-cd marker to cpu_to_as
        when -mno-code-density is given.
        * config/arc/driver-arc.cc (arc_cpu_to_as): Do not request
        -mcode-density from the assembler when the marker is present.

Signed-off-by: Luis Silva <luis.silva@globalfoundries.com>

[luismgsilva]

This patch fixes the original issue reported in #253