[PATCH, i386]: PR target/14631: Fix pinsrw instruction (take 2)

[Uros Bizjak]

Richard Henderson wrote:

>> if (REG_P (operands[2]))
>>   operands[2] = gen_lowpart_SUBREG (HImode, operands[2]);
>>    
>>
>
>No, plain gen_lowpart is correct.  See below.
>
>What you want for the register case is to use %k2, which will
>force "eax" to be printed instead of "ax".  It has no effect
>on memories at all, so it's still safe to use there.
>
>  
>
Yes, %k2 solves the problems!

Attached patch was boostrapped on pentium4-pc-linux-gnu, regtest is in 
progress. It generates correct code for mem and reg cases.

2004-12-28  Richard Henderson  <rth@redhat.com>
        Uros Bizjak  <uros@kss-loka.si>

    PR target/14631
    * config/i386/i386.c (ix86_expand_builtin): [IX86_BUILTIN_PINSRW,
    IX86_BUILTIN_PINSRW128]: Fix wrong selector range in error message.
    * config/i386/i386.md (mmx_pinsrw, sse2_pinsrw): Fix selector
    handling.
    (*mmx_pinsrw, *sse2_pinsrw): New patterns.
    * config/i386/i386/predicates.md (const_pow2_1_to_8_operand,
    const_pow2_1_to_128_operand): New predicates.
   

Uros.

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: pinsrw.diff
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20041229/7783995c/attachment.ksh>