Re: [AratiD@kpit.com: [ PATCH] : SH Assembler generates incorrect opcode for PCMP instructions]

[Daniel Jacobowitz <drow at mvista dot com>]

On Sun, Sep 22, 2002 at 01:30:52PM -0400, Daniel Jacobowitz wrote:
> This patch has been sitting in my mailbox for eight months; as far as I
> can see the problem still exists.  Should reg_n be initialized inside
> of the loop instead of outside?  Or Arati's fix applied?
> 
> I'd appreciate it if an SH maintainer would comment.

Oops.  Upon re-reading I suspect this patch fixed the problem, and that
the loop in assemble_ppi doesn't do what I think it does.  Am I right,
Joern?

2002-06-17  J"orn Rennecke <joern.rennecke@superh.com>

        * config/tc-sh.c (assemble_ppi): Initialize reg_n.

> 
> ----- Forwarded message from Arati Dikey <AratiD@kpit.com> -----
> 
> Date: Thu, 17 Jan 2002 18:10:09 +0530
> From: "Arati Dikey" <AratiD@kpit.com>
> Subject: [ PATCH] : SH Assembler generates incorrect opcode for PCMP instructions
> To: <binutils@sources.redhat.com>
> 
> 
> 
> Hi !
> 
> The SH assembler generates incorrect opcodes for the parallel PCMP
> instruction.
> Basically, its last nibble is a copy of the previous instruction's
> second nibble
> when working in Big Endian format.
> 
> For example,
> test.s
> 	movs.w @-R5, A0
>       PCMP X0,Y0
> 
>  generates opcodes
> 	f5 70
> 	f8 00 84 05
> 
>  instead of
> 	f8 00 84 00
> 
> The following patch corrects this. I have also verified that it does not
> cause any
> side effect on other DSP instructions.
> 
> Regards,
> Arati Dikey
> 
> 
> 
> --- tc-sh.c.orig	Thu Dec  6 11:34:18 2001
> +++ tc-sh.c	      Thu Jan 17 16:41:28 2002
> @@ -1769,7 +1769,10 @@
>  	  if (field_b)
>  	    as_bad (_("multiple parallel processing specifications"));
>  	  field_b = ((opcode->nibbles[2] << 12) + (opcode->nibbles[3] <<
> 8)
> -		     + (reg_x << 6) + (reg_y << 4) + reg_n);
> +		     + (reg_x << 6) + (reg_y << 4) );
> +
> +	  if (strcmp (opcode->name, "pcmp") != 0)
> +	  	 field_b += reg_n;
>  	  break;
>  	case PDC:
>  	  if (cond)
> 
> 
> 
> ----- End forwarded message -----
> 
> -- 
> Daniel Jacobowitz
> MontaVista Software                         Debian GNU/Linux Developer
> 

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer