Bug#894840: clang: baseline cpu violation on i386
Le 28/11/2018 à 00:01, Fanael Linithien a écrit : On Wed, 4 Apr 2018 22:08:16 +0200 Sylvestre Ledru wrote: As this has been the case for a long time, it shows that it isn't a severe issue. It's not a severe issue because it's easy to fix/work around: just passing -march=i686 to Clang is enough. I was never hit by this problem because I'm in the habit of passing the pertinent -march to the compiler always, so I'm not affected by compiler defaults. Besides, this behavior is the upstream default anyway. Yeah, I was incorrect with your bug. This new bug is because I moved to use llvm-toolchain-7 using a stage2 build method. So, all llvm binaries are impacted. I will work on a fix in the next few days S
Bug#894840: clang: baseline cpu violation on i386
On Wed, 4 Apr 2018 22:08:16 +0200 Sylvestre Ledru wrote: > As this has been the case for a long time, it shows that it isn't a severe > issue. It's not a severe issue because it's easy to fix/work around: just passing -march=i686 to Clang is enough. I was never hit by this problem because I'm in the habit of passing the pertinent -march to the compiler always, so I'm not affected by compiler defaults. Besides, this behavior is the upstream default anyway.
Bug#894840: clang: baseline cpu violation on i386
severity 894840 normal thanks Hello, On 04/04/2018 20:36, Andreas Beckmann wrote: > Package: clang > Version: baseline cpu violation on i386 > Severity: serious > > Hi, > > all clang versions (at least since wheezy) by default generate code for > a cpu that is newer than the baseline cpu on i386. On clang-6.0 the > default is apparently pentium4 which has sse2 (among other features), > but that is not available on the baseline cpu, which is currently i686. Sorry about that but this is by design and I am not planning to work on fixing that... As this has been the case for a long time, it shows that it isn't a severe issue. I think we (Debian) should instead have a discussion on having SSE2 as minimal req for intel-like cpu. Other projects like Firefox or Chrome are making that a requirement too. Cheers, Sylvestre
Bug#894840: clang: baseline cpu violation on i386
Package: clang
Version: baseline cpu violation on i386
Severity: serious
Hi,
all clang versions (at least since wheezy) by default generate code for
a cpu that is newer than the baseline cpu on i386. On clang-6.0 the
default is apparently pentium4 which has sse2 (among other features),
but that is not available on the baseline cpu, which is currently i686.
>From clang on wheezy/i386 (but the result is equivalent on newer releases
and also on amd64 with -m32) - clang generates sse code for some floating
point arithmetic, while gcc generates x87 code:
$ echo 'double fma(double a, double b, double c) { return a * b + c; }' | clang
-O2 -S -xc - -o-
.file "-"
.text
.globl fma
.align 16, 0x90
.type fma,@function
fma:# @fma
# BB#0:
subl$12, %esp
movsd 16(%esp), %xmm0
mulsd 24(%esp), %xmm0
addsd 32(%esp), %xmm0
movsd %xmm0, (%esp)
fldl(%esp)
addl$12, %esp
ret
.Ltmp0:
.size fma, .Ltmp0-fma
.section".note.GNU-stack","",@progbits
$ echo 'double fma(double a, double b, double c) { return a * b + c; }' | gcc
-O2 -S -xc - -o-
.file ""
.text
.p2align 4,,15
.globl fma
.type fma, @function
fma:
.LFB0:
.cfi_startproc
fldl12(%esp)
fmull 4(%esp)
faddl 20(%esp)
ret
.cfi_endproc
.LFE0:
.size fma, .-fma
.ident "GCC: (Debian 4.7.2-5) 4.7.2"
.section.note.GNU-stack,"",@progbits
Andreas
