Re: [macppc, !x86] Unbreak emulators/gsplus

On Mon, May 04 2020, Charlene Wendling <julianaito@posteo.jp> wrote:
> On Sun, 3 May 2020 23:48:54 +0200
> Solene Rapenne wrote:
>
>> Le Fri, 1 May 2020 23:40:59 +0200,
>> Charlene Wendling <julianaito@posteo.jp> a écrit :
>>
>> > Hi,
>> >
>> > > http://build-failures.rhaalovely.net/powerpc/2020-04-09/emulators/gsplus.log
>> >
>> > `__builtin_ppc_mftb()' is not available with clang.
>> >
>> > An easy solution would have been to change the ifdef, but as this is
>> > the same situation with retroarch [0], i preferred to use their
>> > logic. On top of that it should slightly improve the emulation on !
>> > {x86,ppc}.
>> >
>> > I don't know why upstream chose to #define a builtin, so it can't be
>> > applied on archs where it is native (x86). The intent of this patch
>> > is to be upstreamed so i preferred to stay minimal.
>> >
>> > This allows gsplus to be built on macppc, i can run System 6 without
>> > issues, excepted mouse clicks that don't register. The only other
>> > arch i have is amd64, where this diff makes no change.
>> >
>> > Comments/feedback are welcome,
>> >
>> > Charlène.
>> >
>> >
>> > [0] https://marc.info/?l=openbsd-ports-cvs&m=158816954508000&w=2
>> >
>> >
>> > Index: Makefile
>> > ===================================================================
>> > RCS file: /cvs/ports/emulators/gsplus/Makefile,v
>> > retrieving revision 1.1.1.1
>> > diff -u -p -u -p -r1.1.1.1 Makefile
>> > --- Makefile 12 Mar 2020 12:24:08 -0000 1.1.1.1
>> > +++ Makefile 1 May 2020 21:36:37 -0000
>> > @@ -6,6 +6,7 @@ GH_ACCOUNT = digarok
>> > GH_PROJECT = gsplus
>> > GH_COMMIT = 480572054518112647c8fae5d7ea7046a6d6ecfb
>> > DISTNAME = ${GH_PROJECT}-20190816
>> > +REVISION = 0
>> >
>> > CATEGORIES = emulators
>> >
>> > Index: patches/patch-src_engine_c_c
>> > ===================================================================
>> > RCS file: patches/patch-src_engine_c_c
>> > diff -N patches/patch-src_engine_c_c
>> > --- /dev/null 1 Jan 1970 00:00:00 -0000
>> > +++ patches/patch-src_engine_c_c 1 May 2020 21:36:37 -0000
>> > @@ -0,0 +1,24 @@
>> > +$OpenBSD$
>> > +
>> > +Use posix timing to improve portability and avoid a build failure
>> > on +powerpc due to `__builtin_ppc_mftb' being gcc-specific.
>> > +
>> > +Index: src/engine_c.c
>> > +--- src/engine_c.c.orig
>> > ++++ src/engine_c.c
>> > +@@ -884,6 +884,15 @@ void fixed_memory_ptrs_shut() {
>> > +
>> > + #if defined(__i386__) || defined(__x86_64__)
>> > + #include <x86intrin.h>
>> > ++ #elif defined(_POSIX_MONOTONIC_CLOCK)
>> > ++ #include <time.h>
>> > ++ int64_t __rdtsc() {
>> > ++ struct timespec tp = {0};
>> > ++ int64_t timestamp = 0;
>> > ++ if (clock_gettime(CLOCK_MONOTONIC, &tp) == 0)
>> > ++ timestamp = tp.tv_sec * 1000000000 + tp.tv_nsec;
>> > ++ return timestamp;
>> > ++ }
>> > + #elif defined(__powerpc__) || defined(__ppc__)
>> > + #define __rdtsc() __builtin_ppc_mftb()
>> > + #else
>>
>> ok solene@
>>
>> Is gsplus usable on powerpc?
>
> This is slow, actually not fast enough to render sound properly, but
> it's usable.

Is it any faster if you drop the powerpc specific bits here? Trading
a single CPU instruction (__builtin_ppc_mftb) for a syscall
(clock_gettime) probably slows things down a bit.

If the #else fallback works fine, maybe it should be preferred.

--
jca | PGP : 0x1524E7EE / 5135 92C1 AD36 5293 2BDF DDCC 0DFA 74AE 1524 E7EE