[M3devel] loophole/copysign

Mon Jul 5 23:32:44 CEST 2010

Are you saying that conversion of floats is not generating a loophole?  The volatilise idea is overkill, and I don't want to see it contaminate the M3CG interfaces.  Can you point me at the problematic code in CastExpr?

On 5 Jul 2010, at 16:49, Jay K wrote:

> 
> The make_volatile idea seems to work. I agree it isn't great. I only inserted one call to it, for loophole + d_to_s + floattype src # dest.
> Let me know if attached is ok to commit.
> The front end all but needs to change here somehow, as the backend isn't being given much information -- no loophole call at all.
> 
> Thanks,
>  - Jay
> 
> ----------------------------------------
>> From: hosking at cs.purdue.edu
>> Date: Mon, 5 Jul 2010 14:26:29 -0400
>> To: jay.krell at cornell.edu
>> CC: m3devel at elegosoft.com
>> Subject: Re: [M3devel] loophole/copysign
>> 
>> We really don't want to do this. We need to figure out what needs to be generated as gcc trees. Perhaps an anonymous struct will suffice.
>> 
>> 
>> On 5 Jul 2010, at 08:56, Jay K wrote:
>> 
>>> 
>>> nope.
>>> I know this is gross, but how about we addd CG.make_volatile that front end can call very sparingly,
>>> that acts just like generating a call to setjmp or vfork -- in that it makes locals, parameters, temporaries
>>> in current function as volatile?
>>> 
>>> CastExpr.m3 would call it for D_to_S and FloatType[src] # FloatType[dest].
>>> 
>>> It would be available as a desparate measure if we find other problems.
>>> To selectively inhibit optimization a function at a time.
>>> Which, granted, is generally overkill.
>>> 
>>> I'm trying this now..
>>> - Jay
>>> 
>>> ----------------------------------------
>>>> From: jay.krell at cornell.edu
>>>> To: m3devel at elegosoft.com
>>>> Subject: RE: [M3devel] loophole/copysign
>>>> Date: Mon, 5 Jul 2010 11:36:11 +0000
>>>> 
>>>> 
>>>> another idea: let's not use bitfield ref for float/double
>>>> 
>>>> - Jay
>>>> 
>>>> 
>>>> 
>>>> ----------------------------------------
>>>>> From: jay.krell at cornell.edu
>>>>> To: m3devel at elegosoft.com
>>>>> Date: Mon, 5 Jul 2010 11:25:19 +0000
>>>>> Subject: Re: [M3devel] loophole/copysign
>>>>> 
>>>>> 
>>>>> Hm. it seems that it might be important to preserve the "designatorness", like in:
>>>>> 
>>>>> libm3/...RandomReal.m3:
>>>>> 
>>>>> VAR frac, exp: INTEGER; result: LONGREAL;
>>>>> 
>>>>> (* Repack as LONGREAL: *)
>>>>> WITH lr = LOOPHOLE (result, LongRealRep.T) DO
>>>>> lr.sign := 0;
>>>>> lr.exponent := exp;
>>>>> lr.significand0 := Word.Shift (Word.And (frac, 16_7fffffff),
>>>>> -(WordSize - 1 - FractionBits));
>>>>> lr.significand1 := r.integer (min := -16_7fffffff-1, max :=16_7fffffff);
>>>>> END;
>>>>> 
>>>>> 
>>>>> - Jay
>>>>> 
>>>>> ----------------------------------------
>>>>>> From: jay.krell at cornell.edu
>>>>>> To: m3devel at elegosoft.com
>>>>>> Date: Mon, 5 Jul 2010 10:42:57 +0000
>>>>>> Subject: Re: [M3devel] loophole/copysign
>>>>>> 
>>>>>> 
>>>>>> Tony, et. al.. in m3front/src/exprs/CastExpr.m3..what's the difference between a "designator" and a "value"?
>>>>>> 
>>>>>> 
>>>>>> http://www-plan.cs.colorado.edu/diwan/modula3/designators.html
>>>>>> 
>>>>>> An identifier is a writable designator
>>>>>> if it is declared as a variable,
>>>>>> is a VAR or VALUE parameter,
>>>>>> is a local of a TYPECASE
>>>>>> or TRY EXCEPT statement,
>>>>>> or is a WITH local that is bound to a writable designator.
>>>>>> An identifier is a readonly designator if it is
>>>>>> a READONLY parameter,
>>>>>> a local of a FOR statement,
>>>>>> or a WITH local bound to a non-designator or
>>>>>> readonly designator.
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> I guess a designator is what I would think of a "variable" or "read only variable"?
>>>>>> Something that either is "in memory" or can "reasonably" be put there?
>>>>>> 
>>>>>> 
>>>>>> 1 + 2 is not a designator.
>>>>>> 
>>>>>> 
>>>>>> Or, generally, a "variable", but that includes such similar things as parameters, "with variables", "for variables", "TYPECASE vairables", "TRY variables"
>>>>>> 
>>>>>> 
>>>>>> Anything with a name??? (not functions/modules/generics -- "named data")
>>>>>> 
>>>>>> 
>>>>>> Anyway, the next questions include:
>>>>>> 
>>>>>> 
>>>>>> In CastExpr.m3 would it be terrible and/or wrong to treat "designators" the same as "values"?
>>>>>> I realize, probably a deoptimization.
>>>>>> I think this lets the backend work.
>>>>>> 
>>>>>> 
>>>>>> And really, more to the point...shouldn't CastExpr.m3 use cg.loophole far more?
>>>>>> I haven't had much luck with that. I always get the cg stack out of balance or with the wrong types, even though it seems like it should be easy.
>>>>>> 
>>>>>> 
>>>>>> I have more testing to do, but classifying the loophole as V_to_S (value to structure) in place of D_to_S (designator to structure), at least if either side is one of the three float types, seems reasonable and correct, albeit slight deoptimization -- in unsafe code dealing with floating point..should be rare..
>>>>>> 
>>>>>> 
>>>>>> - Jay
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> ----------------------------------------
>>>>>>> From: jay.krell at cornell.edu
>>>>>>> To: m3devel at elegosoft.com
>>>>>>> Date: Mon, 5 Jul 2010 09:24:20 +0000
>>>>>>> Subject: [M3devel] loophole/copysign
>>>>>>> 
>>>>>>> 
>>>>>>> Our codegen is remarkably low level. That is, lower level earlier than C.
>>>>>>> 
>>>>>>> 
>>>>>>> gcc/m3cg -ftree-dump-all
>>>>>>> 
>>>>>>> 
>>>>>>> As early as LongFloat.mc.003t.original, the first file dumped, we have:
>>>>>>> 
>>>>>>> LongFloat__CopySign (M3_CtKayy_x, M3_CtKayy_y)
>>>>>>> {
>>>>>>> xreel M3_CtKayy__result;
>>>>>>> xreel M3_CtKayy_res;
>>>>>>> 
>>>>>>> xreel M3_CtKayy__result;
>>>>>>> xreel M3_CtKayy_res;
>>>>>>> M3_CtKayy_res = M3_CtKayy_x;
>>>>>>> BIT_FIELD_REF = (word_8) ((int_64)
>>>>>>> BIT_FIELD_REF & -129 | (word_64) BIT_FIELD_REF <(int_64) BIT_FIELD_REF , 1, 7> << 7 & 255);
>>>>>>> = M3_CtKayy_res;
>>>>>>> return ;
>>>>>>> }
>>>>>>> 
>>>>>>> compared to C where as test_copysign.c.t69.copyrename3, the last file dumped, we have:
>>>>>>> 
>>>>>>> copy_sign_f (from, to)
>>>>>>> {
>>>>>>> float res;
>>>>>>> float D.1918;
>>>>>>> D.1917;
>>>>>>> struct float_t * from.1;
>>>>>>> struct float_t * res.0;
>>>>>>> 
>>>>>>> :
>>>>>>> res = to_1;
>>>>>>> res.0_4 = (struct float_t *) &res;
>>>>>>> from.1_5 = (struct float_t *) &from;
>>>>>>> D.1917_6 = from.1_5->sign;
>>>>>>> res.0_4->sign = D.1917_6;
>>>>>>> D.1918_7 = res;
>>>>>>> return D.1918_7;
>>>>>>> 
>>>>>>> }
>>>>>>> 
>>>>>>> 
>>>>>>> See, you know, from gcc's point of view, we don't have any records/structs/unions.
>>>>>>> Just integers and offsets from them mostly.
>>>>>>> 
>>>>>>> 
>>>>>>> The right fix is to build up types.
>>>>>>> That way also debugging with gdb will have a chance.
>>>>>>> Perhaps not a small amount of work. But maybe not too bad.
>>>>>>> 
>>>>>>> 
>>>>>>> For now my inclination is in m3front to insert a barrier between the store and the load associated with loopholes.
>>>>>>> At least if one type but not the other is floating point.
>>>>>>> I don't know if that will work, but maybe.
>>>>>>> 
>>>>>>> Or maybe have m3front actually call loophole for this case and again, either a barrier or make the load and/or
>>>>>>> store volatile.
>>>>>>> 
>>>>>>> - Jay
>>>>>>> 
>>>>>> 
>>>>> 
>>>> 
>>> 
>> 
> 		 	   		  <volatile.txt>