SH target - disable @(r0,rm) addressing mode?

[<tm_gccmail at mail dot kloo dot net>]

I've been looking at some fairly abysmal SH4 code generated for the
stress-1.14
testcase for CVS head with  -O2 -m4 -fnew-ra, and here's what I've found.

The basic problem seems to be an overdependency on the dual-register
addressing
mode @(r0,rm). The problem is the first register must be r0, so this leads
to an overusage of r0, which leaves no placement freedom for the
instruction
scheduler. The processor appears to be only single-issuing in large blocks
of code due to this problem.

I'm wondering if we should disable the @(r0,rm) addressing mode except for
special cases such as GOT references. This may sound a little extreme,
but I'm getting tired of fighting the problems associated with this single
register class, and I'd like to solve the problem completely for once.

Any thoughts?

Here's the compiled code for reference:

   ( yr  ^= ((((S[0 * 256 + (((    yl   ) >> (24)) & (0x00ff)) ] + S[1 *
256 + (((    yl   ) >> (16)) & (0x00ff)) ]) ^ S[2 * 256 + (((    yl   ) >>
(8)) & (0x00ff)) ]) + S[3 * 256 + (((    yl   ) & 0x00ff)) ])  ^ P[  1
])) ;  ( yl  ^= ((((S[0 * 256 + (((    yr   ) >> (24)) & (0x00ff)) ] + S[1
* 256 + (((    yr   ) >> (16)) & (0x00ff)) ]) ^ S[2 * 256 + (((    yr
) >> (8)) & (0x00ff)) ]) + S[3 * 256 + (((    yr   ) & 0x00ff)) ])  ^ P[
2 ])) ;


        stream2 = S[((yl >> 24) & 0x00ff)]
        stream1 = S[256 + ((yl >> 16) & 0x00ff))]
        stream3 = S[512 + ((yl >> 8 ) & 0x0fff))]
        stream4 = S[768 + (y1 & 0x00ff)]

      12:       99 f7           mov.w   204
<_ssh_blowfish_encrypt+0x204>,r9    ! 0x400
      14:       25 1a           xor     r1,r5
      16:       98 f6           mov.w   206
<_ssh_blowfish_encrypt+0x206>,r8    ! 0x800
      18:       60 53           mov     r5,r0           <- stream 1
      1a:       40 29           shlr16  r0              <- stream 1
      1c:       6a 53           mov     r5,r10          <- stream 2
      1e:       60 0c           extu.b  r0,r0           <- stream 1
      20:       4a 29           shlr16  r10             <- stream 2
      22:       93 f1           mov.w   208
<_ssh_blowfish_encrypt+0x208>,r3    ! 0xc00
      24:       4a 19           shlr8   r10             <- stream 2
      26:       51 21           mov.l   @(4,r2),r1
      28:       40 08           shll2   r0              <- stream 1
      2a:       30 4c           add     r4,r0           <- stream 1
      2c:       4a 08           shll2   r10
      2e:       2e 02           mov.l   r0,@r14         <- stream 1
      30:       60 a3           mov     r10,r0
      32:       0a 4e           mov.l   @(r0,r4),r10
      34:       60 e6           mov.l   @r14+,r0        <- stream 1
      36:       00 9e           mov.l   @(r0,r9),r0     <- stream 1
      38:       3a 0c           add     r0,r10          <- stream 1
      3a:       60 53           mov     r5,r0           <- stream 3
      3c:       40 19           shlr8   r0              <- stream 3
      3e:       60 0c           extu.b  r0,r0           <- stream 3
      40:       40 08           shll2   r0              <- stream 3
      42:       30 4c           add     r4,r0           <- stream 3
      44:       00 8e           mov.l   @(r0,r8),r0     <- stream 3
      46:       2a 0a           xor     r0,r10          <- stream 3
      48:       60 5c           extu.b  r5,r0           <- stream 4
      4a:       40 08           shll2   r0              <- stream 4
      4c:       30 4c           add     r4,r0           <- stream 4
      4e:       00 3e           mov.l   @(r0,r3),r0     <- stream 4
      50:       3a 0c           add     r0,r10          <- stream 4
      52:       2a 1a           xor     r1,r10
...

Toshi