minor optimization missed on SH

[Schirmer, Hartmut]

Hi,

the following code

-------------------
struct
{
    long  x;
    short y;
    struct
    {
        unsigned a : 12;
        unsigned b :  4;
    } s;
} u;

unsigned test(void)
{
    return u.s.a;
}
-------------------

compiles to 

-------------------
	.file	"bitf.i"
	.data
! GNU C version 3.0 20010611 (prerelease) (sh-hitachi-coff)
!	compiled by GNU C version 2.95.3-4 (cygwin special).
! options passed:  -fpreprocessed -mrelax -m1 -ansi -O -Wall
! -Wwrite-strings -Wno-trigraphs -Wunused -Wno-main -Wundef
! -Wmissing-declarations -Wpointer-arith -ansi -fverbose-asm
! -fno-keep-static-consts -fno-strict-aliasing -fomit-frame-pointer
! options enabled:  -fdefer-pop -fomit-frame-pointer -fthread-jumps
! -fpeephole -ffunction-cse -finline -freg-struct-return -fdelayed-branch
! -fsched-interblock -fsched-spec -fbranch-count-reg -fcommon -fverbose-asm
! -fgnu-linker -fargument-alias -fident -fguess-branch-probability
! -fmath-errno -m1 -mrelax

	.text
	.align 2
	.global	_test
_test:
	mov.l	L2,r1
	mov.l	@(4,r1),r0
	shll16	r0
	shlr16	r0
	shlr2	r0
	rts	
	shlr2	r0
L3:
	.align 2
L2:
	.long	_u
	.comm _u,8
	.ident	"GCC: (GNU) 3.0 20010611 (prerelease)"
-------------------

the sequence

	shll16	r0
	shlr16	r0

clears the upper 16bit of r0. Using

	extu.w	r0

would be faster and shorter.

GCC 2.95.2 produces the same code. Code is identical on
ever optimisation level > 0

Hartmut