Should __attribute__((optimize("align-loops"))) work ? (for x86_64)

Should __attribute__((optimize("align-loops"))) work ? (for x86_64)

[Chris Hall]

The manual suggests that it should be possible to use function 
attributes to set "-falign-loops", "-falign-jumps" etc. on a per 
function basis.

With v7.2.1 I find that:

   __attribute__((optimize("align-loops=32")))

is rejected with a "warning: bad option '-falign-loops=32'"

while:

   __attribute__((optimize("align-loops")))

is accepted.

I can compile with "-falign-loops=32" gcc-command-line option.  This 
appears to align loops (to the requested boundary) for "-O0", "-Og", 
"-O3" and (even) "-Os".

I have a large project, with just a few functions for which 
loop-alignment has any material effect -- for the rest of the project, 
it's a waste of space.



I find that "align-functions" is similarly less than

Re: Should __attribute__((optimize("align-loops=32"))) work ? (for x86_64)

[Chris Hall]

With apologies for the noise...

...with v7.2.1, compiling for recent x86_64, the command line option 
"-falign-loops=32" is accepted and has the desired effect.

However I find that both:

     _Pragma("GCC optimize(\"align-loops=32\")")

and:

     __attribute__((optimize("align-loops=32")))

are rejected with a "warning: bad option '-falign-loops=32'" -- reported 
for each and every affected function.

Should I report this as a bug ?

[I have tried optimize("align-loops=x") for x=0, 1, 8 and 16 -- all are 
rejected the same way.]

Thanks,

Chris

Re: Should __attribute__((optimize("align-loops=32"))) work ? (for x86_64)

[Jonathan Wakely]

On 28 January 2018 at 11:25, Chris Hall wrote:
>
> With apologies for the noise...
>
> ...with v7.2.1, compiling for recent x86_64, the command line option
> "-falign-loops=32" is accepted and has the desired effect.
>
> However I find that both:
>
>     _Pragma("GCC optimize(\"align-loops=32\")")
>
> and:
>
>     __attribute__((optimize("align-loops=32")))
>
> are rejected with a "warning: bad option '-falign-loops=32'" -- reported for
> each and every affected function.
>
> Should I report this as a bug ?

Yes, if you can't find an existing report for it.

Re: Should __attribute__((optimize("align-loops=32"))) work ? (for x86_64)

[Chris Hall]

On 28/01/18 12:46, Jonathan Wakely wrote:
> On 28 January 2018 at 11:25, Chris Hall wrote:
>> ...with v7.2.1, compiling for recent x86_64, the command line option
>> "-falign-loops=32" is accepted and has the desired effect.
>>
>> However I find that both:
>>
>>      _Pragma("GCC optimize(\"align-loops=32\")")
>>
>> and:
>>
>>      __attribute__((optimize("align-loops=32")))
>>
>> are rejected with a "warning: bad option '-falign-loops=32'" -- reported for
>> each and every affected function.
>>
>> Should I report this as a bug ?
> Yes, if you can't find an existing report for it.
OK: I have reported it.

FWIW: playing with "Compiler Explorer" suggests that this was broken by 
v7. Compiling with v6.3 does not generate the spurious warning, but the 
requested loop alignment is ignored. Going back to v5.4, I find the 
optimize("align-loops=32") actually works !

Thanks,

Chris