Re: LWLock optimization for multicore Power machines

On Sat, Mar 25, 2017 at 11:32 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> Alexander Korotkov <a(dot)korotkov(at)postgrespro(dot)ru> writes:
> > I moved PPC implementation of pg_atomic_fetch_mask_add_u32() into
> > port/atomics/arch-ppc.h. I also had to declare pg_atomic_uint32 there to
> > satisfy usage of this type as argument
> > of pg_atomic_fetch_mask_add_u32_impl().
>
> Hm, you did something wrong there, because now I get a bunch of failures:
>
> ccache gcc -Wall -Wmissing-prototypes -Wpointer-arith
> -Wdeclaration-after-statement -Wendif-labels -Wmissing-format-attribute
> -Wformat-security -fno-strict-aliasing -fwrapv -g -O2
> -I../../../../src/include -c -o brin.o brin.c
> In file included from ../../../../src/include/port/atomics.h:123,
> from ../../../../src/include/utils/dsa.h:17,
> from ../../../../src/include/nodes/tidbitmap.h:26,
> from ../../../../src/include/access/genam.h:19,
> from ../../../../src/include/nodes/execnodes.h:17,
> from ../../../../src/include/access/brin.h:14,
> from brin.c:18:
> ../../../../src/include/port/atomics/generic.h:154:3: error: #error "No
> pg_atomic_test_and_set provided"
> ../../../../src/include/port/atomics.h: In function 'pg_atomic_init_flag':
> ../../../../src/include/port/atomics.h:178: warning: implicit declaration
> of function 'pg_atomic_init_flag_impl'
> ../../../../src/include/port/atomics.h: In function
> 'pg_atomic_test_set_flag':
> ../../../../src/include/port/atomics.h:193: warning: implicit declaration
> of function 'pg_atomic_test_set_flag_impl'
> ../../../../src/include/port/atomics.h: In function
> 'pg_atomic_unlocked_test_flag':
> ../../../../src/include/port/atomics.h:208: warning: implicit declaration
> of function 'pg_atomic_unlocked_test_flag_impl'
> ... and so on.
>
> I'm not entirely sure what the intended structure of these header files
> is. Maybe Andres can comment.
>

It seems that on this platform definition of atomics should be provided by
fallback.h. But it doesn't because I already
defined PG_HAVE_ATOMIC_U32_SUPPORT in arch-ppc.h. I think in this case we
shouldn't provide ppc-specific implementation
of pg_atomic_fetch_mask_add_u32(). However, I don't know how to do this
assuming arch-ppc.h is included before compiler-specific headers. Thus, in
arch-ppc.h we don't know yet if we would find implementation of atomics for
this platform. One possible solution is to provide assembly implementation
for all atomics in arch-ppc.h.

------
Alexander Korotkov
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company