--- Comment #21 from rguenth at gcc dot gnu dot org 2009-04-08 08:39
---
Ah, the key is the complex type! Thus, the following testcase works
struct Y {};
struct X {
Y y;
__complex__ float val;
};
struct X __attribute__((noinline))
foo (float *p)
{
struct X x;
__real x.val =
--- Comment #22 from paolo dot carlini at oracle dot com 2009-04-08 09:11
---
Excellent, thanks.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39678
--- Comment #5 from rguenth at gcc dot gnu dot org 2009-04-07 14:51 ---
The expected output is
10 10
10 10
(10,0) (10,0)
(10,0) (10,0)
correct? With -O0 I get
10 10
10 10
(nan,nan) (nan,nan)
(10,0) (10,0)
while for -m32 it is ok. So it appears only the float variant is affected?
--- Comment #7 from rguenth at gcc dot gnu dot org 2009-04-07 15:29 ---
Minor data point (-O1 -g):
movss 4(%rbx), %xmm3
movss (%rbx), %xmm2
movss (%rsp), %xmm1
movss 4(%rsp), %xmm0
call__mulsc3
Breakpoint 1, 0x77667990 in
--- Comment #8 from paolo dot carlini at oracle dot com 2009-04-07 15:32
---
Ok, give me a few minutes, I'm going to dig a bit further...
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39678
--- Comment #9 from rguenth at gcc dot gnu dot org 2009-04-07 15:44 ---
Wait - it's ok.
class binder2nd
: public unary_functiontypename _Operation::first_argument_type,
typename _Operation::result_type
{
protected:
_Operation op;
typename
--- Comment #10 from rguenth at gcc dot gnu dot org 2009-04-07 15:45
---
It seems to be an ABI mismatch in the backend somehow.
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
--- Comment #11 from rguenth at gcc dot gnu dot org 2009-04-07 15:46
---
Created an attachment (id=17603)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=17603action=view)
preprocessed testcase with bind2nd marked noinline
Testcase, look at the compiled piece for the call to
--- Comment #12 from paolo dot carlini at oracle dot com 2009-04-07 15:48
---
You are too fast ;)
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39678
--- Comment #13 from rguenth at gcc dot gnu dot org 2009-04-07 15:57
---
The intel compiler generates
..___tag_value__ZSt7bind2ndISt10multipliesISt7complexIfEES2_ESt9binder2ndIT_ERKS5_RKT0_.69:
#165.5
movq (%rsi), %rax #167.53
--- Comment #14 from rguenth at gcc dot gnu dot org 2009-04-07 16:04
---
With a small testcase it works:
struct Y {};
struct X {
Y y;
float real;
float imag;
};
struct X __attribute__((noinline))
foo (float *p)
{
struct X x;
x.real = p[0];
x.imag = p[1];
return x;
}
--- Comment #15 from rguenth at gcc dot gnu dot org 2009-04-07 16:05
---
So maybe it's a C++ FE issue after all ...
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39678
--- Comment #16 from paolo dot carlini at oracle dot com 2009-04-07 16:24
---
Sorry for the noise, forgot that we don't care any more about 4_1 and 4_2 ;)
Anyway, the issue is apparently *very* old...
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39678
--- Comment #17 from paolo dot carlini at oracle dot com 2009-04-07 16:44
---
I guess I'm going to reduce the big one with 'delta', if you have special
advice about that, just let me know...
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39678
--- Comment #18 from paolo dot carlini at oracle dot com 2009-04-07 19:01
---
Created an attachment (id=17604)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=17604action=view)
A bit smaller
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39678
--- Comment #19 from paolo dot carlini at oracle dot com 2009-04-07 20:47
---
Created an attachment (id=17605)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=17605action=view)
Smaller...
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39678
--- Comment #20 from paolo dot carlini at oracle dot com 2009-04-07 22:04
---
Created an attachment (id=17606)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=17606action=view)
Even closer to minimal (as C++ code) and more correct
--
17 matches
Mail list logo
