[Bug c/45600] gcc generates illegal AVX aligned moves

[Bug c/45600] gcc generates illegal AVX aligned moves

[greened at obbligato dot org]

------- Comment #1 from greened at obbligato dot org  2010-09-08 16:08 -------
Compile with -c -mavx reduced.c.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug c/45600] New: gcc generates illegal AVX aligned moves

[greened at obbligato dot org]

For the attached testcase, gcc generates a vmovapd for the store to
llvm_cbe__24__StackDv_P53.  The latest Intel sde generates an alignment error:

SDE ERROR: ALIGN32 FAILED PC=40048b MEMEA=7ffffff057d0 vmovapd ymmword ptr
[rax], ymm0

It looks like gcc is considering 16-byte aligned data to be suitable for a
256-bit vmovapd, which it isn't.


-- 
           Summary: gcc generates illegal AVX aligned moves
           Product: gcc
           Version: 4.5.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: greened at obbligato dot org
 GCC build triplet: x86_64-unknown-linux-gnu
  GCC host triplet: x86_64-unknown-linux-gnu
GCC target triplet: x86_64-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug c/45600] gcc generates illegal AVX aligned moves

[greened at obbligato dot org]

------- Comment #2 from greened at obbligato dot org  2010-09-08 16:09 -------
Created an attachment (id=21740)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21740&action=view)
Reduced testcase


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug target/45600] gcc generates illegal AVX aligned moves

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2010-09-08 17:39 -------
I think this code is undefined with respect of alignment requirements.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug target/45600] gcc generates illegal AVX aligned moves

[pinskia at gcc dot gnu dot org]

------- Comment #4 from pinskia at gcc dot gnu dot org  2010-09-08 17:43 -------
Yes this is invalid with respect of alignment requirements.

It becomes obvious from the optimized at -O0 on the trunk.

  v4df llvm_cbe_r5585;
  v4df llvm_cbe_r5584;
  struct l_DV1 llvm_cbe__24__StackDv_P53;
  unsigned int * D.3215;
  struct l_DV1 * llvm_cbe__24__StackDv_P53.0;

<bb 2>:
  llvm_cbe__24__StackDv_P53.0_1 = &llvm_cbe__24__StackDv_P53;
  MEM[(v4df *)llvm_cbe__24__StackDv_P53.0_1] = llvm_cbe_r5584_2(D); // requires
v4df alignment
  D.3215_3 = &llvm_cbe__24__StackDv_P53.field1.field5;
  MEM[(struct  *)D.3215_3].data = llvm_cbe_r5585_4(D); // requires v4df
alignment


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug target/45600] gcc generates illegal AVX aligned moves

[greened at obbligato dot org]

------- Comment #5 from greened at obbligato dot org  2010-09-08 18:52 -------
Why is the code undefined?  Can you explain in terms of the original test
source?  I don't immediately see anything undefined there.


-- 

greened at obbligato dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |greened at obbligato dot org
             Status|RESOLVED                    |UNCONFIRMED
         Resolution|INVALID                     |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug target/45600] gcc generates illegal AVX aligned moves

[pinskia at gcc dot gnu dot org]

------- Comment #6 from pinskia at gcc dot gnu dot org  2010-09-08 18:55 -------
The alignment of llvm_cbe__24__StackDv_P53 is only 64bits so you are casting to
a greater aligned type and then dereferencing it.

That being said, the LLVM C back-end produces crazy c code that is also
undefined because of aliasing.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug target/45600] gcc generates illegal AVX aligned moves

[greened at obbligato dot org]

------- Comment #7 from greened at obbligato dot org  2010-09-08 18:58 -------
(In reply to comment #5)
> Why is the code undefined?  Can you explain in terms of the original test
> source?  I don't immediately see anything undefined there.

Ah, the cast from int field5 to v4df?  Yes, that doesn't look right to me.  Is
this what you're talking about?

It seems to me that gcc should emit a warning about this and/or not assume that
&field5 is aligned to 32 bytes, as there is nothing in the source specifying
such alignment.

If it's an illegal program, gcc should at least emit a warning, if not an
error.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug target/45600] gcc generates illegal AVX aligned moves

[greened at obbligato dot org]

------- Comment #8 from greened at obbligato dot org  2010-09-08 18:59 -------
(In reply to comment #6)
> The alignment of llvm_cbe__24__StackDv_P53 is only 64bits so you are casting to
> a greater aligned type and then dereferencing it.

I didn't know that typing something as a vector guaranteed alignment.  I don't
see anything in the gcc manual about that.  Isn't that why we have
__attribute(aligned())__?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug target/45600] gcc generates illegal AVX aligned moves

[pinskia at gcc dot gnu dot org]

------- Comment #9 from pinskia at gcc dot gnu dot org  2010-09-08 19:01 -------
>If it's an illegal program, gcc should at least emit a warning, if not an
error.


It is not an invalid program, it is just undefined at runtime.  There was a
defect report against the C standard asking if a diagnostic can be required for
undefined behavior and it was rejected.  Oh there is a patch to enable warnings
for this case already.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug target/45600] gcc generates illegal AVX aligned moves

[pinskia at gcc dot gnu dot org]

------- Comment #10 from pinskia at gcc dot gnu dot org  2010-09-08 19:01 -------
vector types are naturally aligned just like integer types.  That is they are
aligned on their size.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600

[Bug target/45600] gcc generates illegal AVX aligned moves

[greened at obbligato dot org]

------- Comment #11 from greened at obbligato dot org  2010-09-08 19:16 -------
(In reply to comment #9)
> >If it's an illegal program, gcc should at least emit a warning, if not an
> error.
> 
> 
> It is not an invalid program,

Yes, you are quite right.  I will take this up with the LLVM folks.  :)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45600