Andres Freund writes:
> On 2018-12-01 17:28:54 -0500, Tom Lane wrote:
>> I guess I wasn't precise enough: I meant add -msse2 if on x86 and
>> compiler doesn't take -fexcess-precision=standard.
> Hm, I still don't like that: It'd silently bump the minimum required
> architecture. Like in the case
Hi,
On 2018-12-01 17:28:54 -0500, Tom Lane wrote:
> Andres Freund writes:
> > On 2018-12-01 12:06:36 -0500, Tom Lane wrote:
> >> Or should we go so far as to have configure add -msse2 automatically?
>
> > Probably not - there's several OSs / distributions that do not yet
> > require an SSE2 capa
Andres Freund writes:
> On 2018-12-01 12:06:36 -0500, Tom Lane wrote:
>> I wonder whether we shouldn't remove the clang aspect of the test
>> you added, ie just make it read "if on x86, you must have either
>> -fexcess-precision=standard or -msse2".
> That sounds reasonable.
>> Or should we go s
Hi,m
On 2018-12-01 12:06:36 -0500, Tom Lane wrote:
> I wonder whether we shouldn't remove the clang aspect of the test
> you added, ie just make it read "if on x86, you must have either
> -fexcess-precision=standard or -msse2".
That sounds reasonable.
> Or should we go so far as to have configu
I wrote:
> Looking at the buildfarm, our only extant member that is on x86, and
> is not using clang, and doesn't have -fexcess-precision=standard, is
> dromedary which is using Apple's old toolchain. So the fact that it
> isn't showing the problem isn't very good evidence about how widespread
> t
Andres Freund writes:
> On 2018-09-04 17:51:30 -0700, Andres Freund wrote:
>> My current proposal is thus to do add a check that does
>> #if defined(__clang__) && defined(__i386__) && !defined(__SSE2_MATH__)
>> something-that-fails
>> #endif
>> in an autoconf test, and have configure complain if t
Hi,
On 2018-09-13 14:29:59 -0700, Andres Freund wrote:
> On 2018-09-04 17:51:30 -0700, Andres Freund wrote:
> > My current proposal is thus to do add a check that does
> > #if defined(__clang__) && defined(__i386__) && !defined(__SSE2_MATH__)
> > something-that-fails
> > #endif
> > in an autoconf
On 2018-09-04 17:51:30 -0700, Andres Freund wrote:
> My current proposal is thus to do add a check that does
> #if defined(__clang__) && defined(__i386__) && !defined(__SSE2_MATH__)
> something-that-fails
> #endif
> in an autoconf test, and have configure complain if that
> fails. Something roughly
Andrew Gierth writes:
> The problem is that if we're relying on -fexcess-precision=standard
> semantics in places besides infinity checks, then we won't get those
> semantics on clang/i386/no-sse2 since it has no comparable option. (What
> are we doing about compilers for x86-32 other than clang a
> "Tom" == Tom Lane writes:
>> At the very minimum, I believe FreeBSD project policy would require
>> that the i386 packages for postgresql be built to run without SSE2.
Tom> Well, can we tell them they have to use gcc to compile, or will
Tom> that be against their policy too?
That's no
Andrew Gierth writes:
> "Tom" == Tom Lane writes:
> Tom> If you wanted to argue that the set of people who still want to
> Tom> run PG on pre-SSE2 hardware is the empty set, that'd be an easier
> Tom> sell really.
> At the very minimum, I believe FreeBSD project policy would require that
> th
> "Peter" == Peter Eisentraut writes:
> On 05/09/2018 18:42, Andres Freund wrote:
>> Realistically we're going to be running into old versions of clang
>> for a long time. And the importance of running i386 without SSE2
>> surely isn't increasing. So I don't really see an urgent need to d
> "Tom" == Tom Lane writes:
Tom> If you wanted to argue that the set of people who still want to
Tom> run PG on pre-SSE2 hardware is the empty set, that'd be an easier
Tom> sell really.
At the very minimum, I believe FreeBSD project policy would require that
the i386 packages for postgres
On 2018-09-05 14:08:11 -0400, Tom Lane wrote:
> Andres Freund writes:
> > On 2018-09-05 10:05:26 -0400, Tom Lane wrote:
> >> One thought is that maybe we should provide a way to override this,
> >> in case somebody really can't or doesn't want to use -msse2, and
> >> is willing to put up with plat
Andres Freund writes:
> On 2018-09-05 10:05:26 -0400, Tom Lane wrote:
>> One thought is that maybe we should provide a way to override this,
>> in case somebody really can't or doesn't want to use -msse2, and
>> is willing to put up with platform-dependent float behavior.
> IDK, people that are c
Andres Freund writes:
> On 2018-09-05 18:55:34 +0200, Peter Eisentraut wrote:
>> Another option perhaps is to let this be and accept it as alternative
>> floating point behavior. We already have some of those.
> -many. We'd directly violate our own error rules.
I think that just from a regress
On 2018-09-05 18:55:34 +0200, Peter Eisentraut wrote:
> On 05/09/2018 18:42, Andres Freund wrote:
> > Realistically we're going to be running into old versions of clang for a
> > long time. And the importance of running i386 without SSE2 surely isn't
> > increasing. So I don't really see an urgen
On 05/09/2018 18:42, Andres Freund wrote:
> Realistically we're going to be running into old versions of clang for a
> long time. And the importance of running i386 without SSE2 surely isn't
> increasing. So I don't really see an urgent need to do anything about
> it. And if it gets fixed, and we
Hi,
On 2018-09-05 10:05:26 -0400, Tom Lane wrote:
> Peter Eisentraut writes:
> > On 05/09/2018 02:51, Andres Freund wrote:
> >> My current proposal is thus to do add a check that does
> >> #if defined(__clang__) && defined(__i386__) && !defined(__SSE2_MATH__)
> >> something-that-fails
> >> #endif
Peter Eisentraut writes:
> On 05/09/2018 02:51, Andres Freund wrote:
>> My current proposal is thus to do add a check that does
>> #if defined(__clang__) && defined(__i386__) && !defined(__SSE2_MATH__)
>> something-that-fails
>> #endif
>> in an autoconf test, and have configure complain if that fa
On 05/09/2018 02:51, Andres Freund wrote:
> My current proposal is thus to do add a check that does
> #if defined(__clang__) && defined(__i386__) && !defined(__SSE2_MATH__)
> something-that-fails
> #endif
> in an autoconf test, and have configure complain if that
> fails. Something roughly along th
On Tue, 4 Sep 2018 17:51:30 -0700
Andres Freund wrote:
> #endif
> in an autoconf test, and have configure complain if that
> fails. Something roughly along the lines of
> "Compiling PostgreSQL with clang, on 32bit x86, requires SSE2
> support. Use -msse2 or use gcc."
Adding CFLAGS=-msse2 to con
On 2018-09-05 01:47:52 +0100, Andrew Gierth wrote:
> > "Tom" == Tom Lane writes:
>
> >> I kinda wonder if we should add -mno-x87 or such in configure when
> >> we detect clang, obviously it doesn't deal correctly with this.
>
> Tom> Seems worth looking into, but what happens if someone tr
> "Tom" == Tom Lane writes:
>> I kinda wonder if we should add -mno-x87 or such in configure when
>> we detect clang, obviously it doesn't deal correctly with this.
Tom> Seems worth looking into, but what happens if someone tries to
Tom> compile for x87 hardware? Or do we care anymore?
On 2018-09-04 19:02:02 -0400, Tom Lane wrote:
> Andres Freund writes:
> > ... Note that we've previously encountered similar issues
> > on gcc, which is why we've tried to force gcc's hand with
> > -fexcess-precision=standard.
>
> Right. Annoying that clang doesn't have that. We can't realistic
Andres Freund writes:
> ... Note that we've previously encountered similar issues
> on gcc, which is why we've tried to force gcc's hand with
> -fexcess-precision=standard.
Right. Annoying that clang doesn't have that. We can't realistically
program around an issue that might or might not show
Hi,
On 2018-09-04 15:16:27 -0700, Andres Freund wrote:
> I think this is pretty clearly a C99 violation by clang (note how gcc
> automatically enables -fexcess-precision=standard in C99 mode). I'll
> report something, but I've little hope this getting fixed quickly - and
> it'll definitely not ge
On 2018-09-04 15:16:27 -0700, Andres Freund wrote:
> I'm now looking at how it comes that clang on linux doesn't default to
> x87 math, but freebsd does.
Oh well. I present you, without comments, the following:
switch (Triple.getOS()) {
case llvm::Triple::FreeBSD:
case llvm::Triple::NetBSD:
On 2018-09-04 17:46:29 -0400, Tom Lane wrote:
> Andrew Gierth writes:
> > The reason it behaves oddly is this: on i387 FPU (and NOT on arm32 or on
> > 32-bit i386 with a modern architecture specified to the compiler), the
> > result of 1e200 * 1e180 is not in fact infinite, because it fits in an
>
> "Tom" == Tom Lane writes:
>> The reason it behaves oddly is this: on i387 FPU (and NOT on arm32
>> or on 32-bit i386 with a modern architecture specified to the
>> compiler), the result of 1e200 * 1e180 is not in fact infinite,
>> because it fits in an 80-bit long double. So __builtin_i
> "Andrew" == Andrew Gierth writes:
Andrew> Specifying a recent microarch makes it use 64-bit FP registers
Andrew> rather than 80-bit ones:
Andrew> cc -O2 -m32 -march=skylake flttst.c && ./a.out
Andrew> r1 = 1, r2 = 1
where "recent" means "since about 2000 or so":
cc -O2 -m32 -march=pe
Andrew Gierth writes:
> The reason it behaves oddly is this: on i387 FPU (and NOT on arm32 or on
> 32-bit i386 with a modern architecture specified to the compiler), the
> result of 1e200 * 1e180 is not in fact infinite, because it fits in an
> 80-bit long double. So __builtin_isinf reports that i
> "Andrew" == Andrew Gierth writes:
> "Andres" == Andres Freund writes:
Andres> Thomas and I are sitting in a cafe and are trying to figure out
Andres> what's going on...
Andrew> I have a standalone test case:
Andrew> #include
Andrew> #include
Andrew> int main(int argc, char *
> "Andres" == Andres Freund writes:
Andres> Thomas and I are sitting in a cafe and are trying to figure out
Andres> what's going on...
I have a standalone test case:
#include
#include
int main(int argc, char **argv)
{
double d1 = (argc ? 1e180 : 0);
double d2 = (argv ? 1e200 :
On 2018-09-04 16:13:45 -0400, Tom Lane wrote:
> Victor Wagner writes:
> > Tom Lane пишет:
> >> (I was testing PG HEAD, not the 11 branch, but I don't see a reason
> >> to think that'd make a difference.)
>
> > Alas, it does. First thing I've done after discovering this bug, it is
> > to look if
> "Victor" == Victor Wagner writes:
Victor> Do you reproducing problem in REL_11_STABLE? It doesn't exist
Victor> in master.
Oh, I missed that. It is reproducible on REL_11_STABLE with clang 3.9.1,
I'll dig into it more.
--
Andrew (irc:RhodiumToad)
Victor Wagner writes:
> Tom Lane пишет:
>> (I was testing PG HEAD, not the 11 branch, but I don't see a reason
>> to think that'd make a difference.)
> Alas, it does. First thing I've done after discovering this bug, it is
> to look if it exists in master. And master passes this test on the same
On Tue, Sep 4, 2018 at 12:59 PM Victor Wagner wrote:
>
> В Tue, 04 Sep 2018 15:48:24 -0400
> Tom Lane пишет:
>
>
> > I tried to reproduce the problem, without success, on a few different
> > FreeBSD images I had laying about:
> >
> > FreeBSD 11.0/x86_64, clang version 3.8.0
> > (this confir
В Tue, 04 Sep 2018 15:48:24 -0400
Tom Lane пишет:
> I tried to reproduce the problem, without success, on a few different
> FreeBSD images I had laying about:
>
> FreeBSD 11.0/x86_64, clang version 3.8.0
> (this confirms OP's report that x86_64 is OK)
>
> FreeBSD 10.3/ppc, gcc 4.2.1
>
>
В Tue, 4 Sep 2018 11:06:56 -0700
Thomas Munro пишет:
> On Tue, Sep 4, 2018 at 9:39 AM Andrew Gierth
> wrote:
> >
> > > "Andres" == Andres Freund writes:
> >
> > >> However, this commit broke float8 test on 32-bit FreeBSD 11 with
> > >> clang 3.8.0 compiler. Regressions.diff follows:
Thomas Munro writes:
> On Tue, Sep 4, 2018 at 9:39 AM Andrew Gierth
> wrote:
>> I also confirmed that without #define isinf(x) __builtin_isinf(x), on
>> both 32bit and 64bit fbsd isinf() compiles as a function call, so the
>> OP's proposed change would not be desirable.
Presumably what we are lo
On Tue, Sep 4, 2018 at 9:39 AM Andrew Gierth
wrote:
>
> > "Andres" == Andres Freund writes:
>
> >> However, this commit broke float8 test on 32-bit FreeBSD 11 with
> >> clang 3.8.0 compiler. Regressions.diff follows:
>
> Andres> Does this happen with a newer clang version too?
>
> float8 t
> "Andres" == Andres Freund writes:
>> However, this commit broke float8 test on 32-bit FreeBSD 11 with
>> clang 3.8.0 compiler. Regressions.diff follows:
Andres> Does this happen with a newer clang version too?
float8 test (and all other tests) passes for me on clang 3.9.1 on fbsd11
on
On September 4, 2018 6:16:24 AM PDT, Victor Wagner wrote:
>
>Collegues,
>
>Few days ago commit
>
> 1f349aa7d9a6633e87db071390c73a39ac279ba4
>
>Fix 8a934d67 for libc++ and make more include order resistant.
>
>was introduced into branch REL_11_STABLE.
>It seems that it deals with GLIBC-based s
Collegues,
Few days ago commit
1f349aa7d9a6633e87db071390c73a39ac279ba4
Fix 8a934d67 for libc++ and make more include order resistant.
was introduced into branch REL_11_STABLE.
It seems that it deals with GLIBC-based systems (i.e Linux and Hurd)
only, and shouldn't affect systems with non-
45 matches
Mail list logo
