Re: [Fwd: [Bug rtl-optimization/33927] New: replace_read in dse.c could handle cases where GET_MODE_CLASS (read_mode) != GET_MODE_CLASS (store_mode) (and the size is the same)]

[Kenneth Zadeck <zadeck at naturalbridge dot com>]

Richard Sandiford wrote:
> Kenneth Zadeck <zadeck@naturalbridge.com> writes:
>   
>> You should make sure that iant is kept aware of the progress or lack
>> of.  Sometimes this kind of thing can fall into a pit because no one who
>> cares is watching. 
>>     
>
> Right. ;)
>
>   
>>>> I was wondering if you had any thoughts on this bugzilla.  I have never
>>>> looked at vector registers and really have no idea what is involved here.
>>>>     
>>>>         
>>> That's a coincidence... I was working on this code for another reason;
>>> see the patch that I'm about to post.  (I don't want to become the defacto
>>> owner of this code though!)
>>>   
>>>       
>> I promise to only ask mode related questions. The rest i will continue
>> to handle.  There really are only a few people who know the edge case of
>> how modes are used and what they mean.
>>
>> No patch was included in your mail. 
>>     
>
> Sorry, I meant the patch that I'm about to post to gcc-patches.
> (Now sent, subject "RFA: DSE replace_read revisited".)
>
> Richard
>   
So i did try your patch on my x86-{32,64} ppc-32, and ia-64. 
you patch as posted seems to CLEAR one failure on the ia-64 but
otherwise causes no regressions.

 Executing on host: /home/zadeck/gbBaseline/gcc/xgcc
-B/home/zadeck/gbBaseline/gcc/
/home/zadeck/gccBaseline/gcc/testsuite/gcc.dg/pr30643.c   -O2
-fno-show-column -S  -o pr30643.s    (timeout = 300)
PASS: gcc.dg/pr30643.c (test for excess errors)
FAIL: gcc.dg/pr30643.c scan-assembler-not undefined


I will rerun all of this if pinskia once pinskia gives the all clear
with what ever it takes to clear his issue. 

kenny