Re: SH: remove -mbigtable option (PR c/4516)

[Joern Rennecke <amylaar at onetel dot net dot uk>]

> 2001-10-12  NIIBE Yutaka  <gniibe@m17n.org>
> 
> 	* doc/invoke.texi (Option Summary: SH Options): Remove -mbigtable.
> 
> 	* config/sh/sh.md (*casesi_worker): Handle 0..65535.  Set attribute
> 	"length" with shl_casesi_worker_length.
> 
> 	* config/sh/sh.h (BIGTABLE_BIT, TARGET_BIGTABLE): Removed.
> 	(TARGET_SWITCHES): Remove -mbigtable option.
> 	(CASE_VECTOR_MODE): Default to SImode, QImode when optimized
> 	(ASM_OUTPUT_ADDR_VEC_ELT): Always emit .long.
> 	(CASE_VECTOR_SHORTEN_MODE): Handle 0..65535.
> 	(optimized_size): Declared.
> 
> 	* config/sh/sh.c (shl_casesi_worker_length): New function.
> 	* config/sh/sh-protos.h (shl_casesi_worker_length): New function.

It's almost there.  Except I forgot to mention that the branch shortening
won't look at the length attribute a second time unless you explicitly
mention (pc) or (match_dup <num>) somewhere in it.  A match_dup in this
context expects an label or a label_ref, so you could use (match_dup 2).
You can look at the default definition of the length attribute for an
example of how to make this be a no-op.
When you use (pc), this will be translated into a call to
insn_current_reference_address, which is a bit on the costly side.  OTOH,
you could avoid ever executing the call - and even have it optimized out -
by first using a test that is obviously constant for the C compiler, but
opaque for genattrtab, like (eq (symbol_ref "0") (const_int 0)).

Whatever you do, it probably deserves a comment.

> + /* Function to be used in the length attribute of the casesi_worker instruction */

Please add a comment to describe the return value.  When I first looked at
this, I thought it was wrong, till I looked at the caller.
> + int
> + shl_casesi_worker_length (insn)