true. Thx!
Sun
On Wed, Jun 29, 2011 at 7:39 AM, Xiaohua Zhang <xiaohua_zh...@yahoo.com> wrote:
> Hi Sun,
> It is target independent, all the targets need to get the right type of the
> expression in order to find if the struct is returned in register and which
> register contains the return value. This patch only fixed the problem that
> the compiler mistakenly used struct type as return type, instead of field
> type, for field access expression like "m.f2". Once the right type is found,
> the target dependent part of the compiler will do the rest.
> Thanks,
> -Xiaohua
> ________________________________
> From: Sun Chan <sun.c...@gmail.com>
> To: Jian-Xin Lai <laij...@gmail.com>
> Cc: Xiaohua Zhang <xiaohua_zh...@yahoo.com>;
> open64-devel@lists.sourceforge.net
> Sent: Tue, June 28, 2011 3:50:15 PM
> Subject: Re: [Open64-devel] code review request for bug 799
>
> sorry for missing this. This change is not target independent is it?
> Sun
>
> On Tue, Jun 28, 2011 at 11:08 AM, Jian-Xin Lai <laij...@gmail.com> wrote:
>> This patch looks fine to me. Please go ahead.
>>
>> 2011/6/25 Xiaohua Zhang <xiaohua_zh...@yahoo.com>:
>>> Hi,
>>>
>>> Could a gatekeeper help to review the fix for bug 799?
>>>
>>> Thanks,
>>> -Xiaohua
>>>
>>> The small test case is:
>>>
>>> #include <stdio.h>
>>> int rval = 0;
>>> typedef struct S { double i; } S;
>>> S f(void);
>>> typedef struct {
>>> int f1;
>>> struct S f2;
>>> } S1;
>>> S1 m;
>>> struct S t = { 0.0 };
>>> int main(void)
>>> {
>>> t.i = 12.345;
>>> m.f2 = f(); // return value should be in xmm0
>>> if (m.f2.i != t.i) {
>>> rval = -1;
>>> printf("Failed!\n");
>>> }
>>> return rval;
>>> }
>>> struct S f(void)
>>> {
>>> return t;
>>> }
>>>
>>> The WHIRL tree for "m.f2 = f()" looks like:
>>>
>>> MCALL 126 <1,53,f> # flags 0x7e {line: 1/20}
>>> MMLDID -1 <1,49,.preg_return_val> T<53,S,8>
>>> MSTID 8 <1,54,m> T<55,.anonymous.1,8> <field_id:2> {line: 0/0}
>>>
>>> After the RETURN_VAL & MLDID/MSTID lowering:
>>>
>>> MCALL 126 <1,53,f> # flags 0x7e {line: 1/20}
>>> I8I8LDID 1 <1,5,.preg_I8> T<5,.predef_I8,8> # $r1
>>> I8STID 8 <1,54,m> T<5,.predef_I8,8> {line: 0/0}
>>> F8F8LDID 17 <1,11,.preg_F8> T<11,.predef_F8,8> # $f0
>>> F8STID 16 <1,54,m> T<11,.predef_F8,8> {line: 0/0}
>>>
>>> The return value (S) is returned by two register: $r1 and $f0.
>>> By the ABI, the f()'s return value (type struct S) should be returned by
>>> one
>>> floating register $f0.
>>> The problem is in lower_return_mstid(), which takes MSTID's type (S1) as
>>> return type, didn't consider the field (f2) it accessed. So it used the
>>> wrong type (S1)to get return info, and the return value for S2 is by
>>> register $r1 and $f0.
>>>
>>> The fix is to get the accessed field type as return type.
>>>
>>>
>>>
>>>
>>>
>>> ----------------------
>>>
>>> Index: be/com/wn_lower.cxx
>>> ===================================================================
>>> --- be/com/wn_lower.cxx (revision 3659)
>>> +++ be/com/wn_lower.cxx (working copy)
>>> @@ -7239,7 +7239,12 @@
>>> ST *preg_st;
>>> WN *n_rhs;
>>> WN *wn = NULL; // init to prevent upward-exposed use
>>> - RETURN_INFO return_info = Get_Return_Info(WN_ty(tree),
>>> Complex_Not_Simulated
>>> + TY_IDX ty_idx = WN_ty(tree);
>>> +
>>> + if (WN_field_id (tree) != 0)
>>> + ty_idx = get_field_type (ty_idx, WN_field_id (tree));
>>> +
>>> + RETURN_INFO return_info = Get_Return_Info(ty_idx,
>>> Complex_Not_Simulated
>>> #ifdef TARG_X8664
>>> , last_call_ff2c_abi
>>> #endif
>>> @@ -7260,7 +7265,7 @@
>>> #endif
>>> WN *awn = WN_CreateLda(OPR_LDA, Pointer_Mtype, MTYPE_V,
>>> WN_store_offset(tree),
>>> - Make_Pointer_Type(WN_ty(tree)),
>>> WN_st_idx(tree));
>>> + Make_Pointer_Type(ty_idx), WN_st_idx(tree));
>>> awn = lower_expr(block, awn, actions);
>>> WN *n_call = add_fake_parm(call, awn, WN_ty(awn));
>>> WN_DELETE_FromBlock(block, call);
>>> @@ -7277,9 +7282,9 @@
>>> mtype = RETURN_INFO_mtype(return_info, i);
>>> desc = mtype;
>>> #ifdef KEY
>>> - if (MTYPE_byte_size(mtype) > TY_size(WN_ty(tree)) &&
>>> !MTYPE_float(mtype)){
>>> + if (MTYPE_byte_size(mtype) > TY_size(ty_idx) &&
>>> !MTYPE_float(mtype)){
>>> desc = Mtype_AlignmentClass(1, MTYPE_type_class(mtype));
>>> - while (MTYPE_byte_size(desc) < TY_size(WN_ty(tree)))
>>> + while (MTYPE_byte_size(desc) < TY_size(ty_idx))
>>> desc = Mtype_next_alignment(desc);
>>> }
>>> #endif
>>> @@ -7287,7 +7292,7 @@
>>>
>>> #if defined(TARG_PPC32)
>>> if (preg_st == Int_Preg) {
>>> - UINT rem = TY_size(WN_ty(tree)) - i * 4;
>>> + UINT rem = TY_size(ty_idx) - i * 4;
>>> if (1 == rem) desc = MTYPE_I1;
>>> if (2 == rem) desc = MTYPE_I2;
>>> }
>>>
>>>
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> All the data continuously generated in your IT infrastructure contains a
>>> definitive record of customers, application performance, security
>>> threats, fraudulent activity and more. Splunk takes this data and makes
>>> sense of it. Business sense. IT sense. Common sense..
>>> http://p.sf.net/sfu/splunk-d2d-c1
>>> _______________________________________________
>>> Open64-devel mailing list
>>> Open64-devel@lists.sourceforge.net
>>> https://lists.sourceforge.net/lists/listinfo/open64-devel
>>>
>>>
>>
>>
>>
>> --
>> Regards,
>> Lai Jian-Xin
>>
>>
>> ------------------------------------------------------------------------------
>> All of the data generated in your IT infrastructure is seriously valuable.
>> Why? It contains a definitive record of application performance, security
>> threats, fraudulent activity, and more. Splunk takes this data and makes
>> sense of it. IT sense. And common sense.
>> http://p.sf.net/sfu/splunk-d2d-c2
>> _______________________________________________
>> Open64-devel mailing list
>> Open64-devel@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/open64-devel
>>
>
------------------------------------------------------------------------------
All of the data generated in your IT infrastructure is seriously valuable.
Why? It contains a definitive record of application performance, security
threats, fraudulent activity, and more. Splunk takes this data and makes
sense of it. IT sense. And common sense.
http://p.sf.net/sfu/splunk-d2d-c2
_______________________________________________
Open64-devel mailing list
Open64-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/open64-devel
