* [PATCH v2] powerpc: provide __bswapdi2
2013-05-13 7:09 ` Michael Neuling
@ 2013-05-13 7:20 ` Michael Neuling
2013-05-13 7:33 ` [PATCH] " Gabriel Paubert
` (2 subsequent siblings)
3 siblings, 0 replies; 18+ messages in thread
From: Michael Neuling @ 2013-05-13 7:20 UTC (permalink / raw)
To: benh, David Woodhouse; +Cc: linuxppc-dev, viro
From: David Woodhouse <David.Woodhouse@intel.com>
Some versions of GCC apparently expect this to be provided by libgcc.
Updates from Mikey to fix 32 bit version and adding "r" to registers.
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Michael Neuling <mikey@neuling.org>
diff --git a/arch/powerpc/kernel/misc_32.S b/arch/powerpc/kernel/misc_32.S
index 19e096b..3bcaa7ee 100644
--- a/arch/powerpc/kernel/misc_32.S
+++ b/arch/powerpc/kernel/misc_32.S
@@ -657,6 +657,17 @@ _GLOBAL(__ucmpdi2)
li r3,2
blr
+_GLOBAL(__bswapdi2)
+ rotlwi r9,r4,8
+ rotlwi r10,r3,8
+ rlwimi r9,r4,24,0,7
+ rlwimi r10,r3,24,0,7
+ rlwimi r9,r4,24,16,23
+ rlwimi r10,r3,24,16,23
+ mr r4,r10
+ mr r3,r9
+ blr
+
_GLOBAL(abs)
srawi r4,r3,31
xor r3,r3,r4
diff --git a/arch/powerpc/kernel/misc_64.S b/arch/powerpc/kernel/misc_64.S
index 5cfa800..637bd82 100644
--- a/arch/powerpc/kernel/misc_64.S
+++ b/arch/powerpc/kernel/misc_64.S
@@ -234,6 +234,18 @@ _GLOBAL(__flush_dcache_icache)
isync
blr
+_GLOBAL(__bswapdi2)
+ srdi r8,r3,32
+ rlwinm r7,r3,8,0xffffffff
+ rlwimi r7,r3,24,0,7
+ rlwinm r9,r8,8,0xffffffff
+ rlwimi r7,r3,24,16,23
+ rlwimi r9,r8,24,0,7
+ rlwimi r9,r8,24,16,23
+ sldi r7,r7,32
+ or r7,r7,r9
+ mr r3,r7
+ blr
#if defined(CONFIG_PPC_PMAC) || defined(CONFIG_PPC_MAPLE)
/*
diff --git a/arch/powerpc/kernel/ppc_ksyms.c b/arch/powerpc/kernel/ppc_ksyms.c
index 78b8766..c296665 100644
--- a/arch/powerpc/kernel/ppc_ksyms.c
+++ b/arch/powerpc/kernel/ppc_ksyms.c
@@ -143,7 +143,8 @@ EXPORT_SYMBOL(__lshrdi3);
int __ucmpdi2(unsigned long long, unsigned long long);
EXPORT_SYMBOL(__ucmpdi2);
#endif
-
+long long __bswapdi2(long long);
+EXPORT_SYMBOL(__bswapdi2);
EXPORT_SYMBOL(memcpy);
EXPORT_SYMBOL(memset);
EXPORT_SYMBOL(memmove);
^ permalink raw reply related [flat|nested] 18+ messages in thread* Re: [PATCH] powerpc: provide __bswapdi2
2013-05-13 7:09 ` Michael Neuling
2013-05-13 7:20 ` [PATCH v2] " Michael Neuling
@ 2013-05-13 7:33 ` Gabriel Paubert
2013-05-13 10:23 ` [PATCH v3] " Michael Neuling
2013-05-13 10:33 ` [PATCH] " David Woodhouse
2013-05-14 1:09 ` Stephen Rothwell
3 siblings, 1 reply; 18+ messages in thread
From: Gabriel Paubert @ 2013-05-13 7:33 UTC (permalink / raw)
To: Michael Neuling; +Cc: linuxppc-dev, David Woodhouse, viro
On Mon, May 13, 2013 at 05:09:59PM +1000, Michael Neuling wrote:
> David Woodhouse <dwmw2@infradead.org> wrote:
>
> > From: David Woodhouse <David.Woodhouse@intel.com>
> >
> > Some versions of GCC apparently expect this to be provided by libgcc.
> >
> > Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
> > ---
> > Untested.
> >
> > diff --git a/arch/powerpc/kernel/misc_32.S b/arch/powerpc/kernel/misc_32.S
> > index 19e096b..f077dc2 100644
> > --- a/arch/powerpc/kernel/misc_32.S
> > +++ b/arch/powerpc/kernel/misc_32.S
> > @@ -657,6 +657,17 @@ _GLOBAL(__ucmpdi2)
> > li r3,2
> > blr
> >
> > +_GLOBAL(__bswapdi2)
> > + rlwinm 10,4,8,0xffffffff
> > + rlwinm 11,3,8,0xffffffff
> > + rlwimi 10,4,24,0,7
> > + rlwimi 11,3,24,0,7
> > + rlwimi 10,4,24,16,23
> > + rlwimi 11,3,24,16,23
> > + mr 4,11
> > + mr 3,10
> > + blr
> > +
>
> This doesn't work for me but the below does:
>
> _GLOBAL(__bswapdi2)
> rotlwi r9,r4,8
> rotlwi r10,r3,8
> rlwimi r9,r4,24,0,7
> rlwimi r10,r3,24,0,7
> rlwimi r9,r4,24,16,23
> rlwimi r10,r3,24,16,23
> mr r4,r10
> mr r3,r9
> blr
>
Actually, I'd swap the two mr instructions to never
have an instruction that uses the result from the
previous one.
> stolen from GCC -02 output of:
> unsigned long long __bswapdi2(unsigned long long x)
> {
> return ((x & 0x00000000000000ffULL) << 56) |
> ((x & 0x000000000000ff00ULL) << 40) |
> ((x & 0x0000000000ff0000ULL) << 24) |
> ((x & 0x00000000ff000000ULL) << 8) |
> ((x & 0x000000ff00000000ULL) >> 8) |
> ((x & 0x0000ff0000000000ULL) >> 24) |
> ((x & 0x00ff000000000000ULL) >> 40) |
> ((x & 0xff00000000000000ULL) >> 56);
> }
>
> > _GLOBAL(abs)
> > srawi r4,r3,31
> > xor r3,r3,r4
> > diff --git a/arch/powerpc/kernel/misc_64.S b/arch/powerpc/kernel/misc_64.S
> > index 5cfa800..3b2e6e8 100644
> > --- a/arch/powerpc/kernel/misc_64.S
> > +++ b/arch/powerpc/kernel/misc_64.S
> > @@ -234,6 +234,18 @@ _GLOBAL(__flush_dcache_icache)
> > isync
> > blr
> >
> > +_GLOBAL(__bswapdi2)
> > + srdi 8,3,32
> > + rlwinm 7,3,8,0xffffffff
> > + rlwimi 7,3,24,0,7
> > + rlwinm 9,8,8,0xffffffff
> > + rlwimi 7,3,24,16,23
> > + rlwimi 9,8,24,0,7
> > + rlwimi 9,8,24,16,23
> > + sldi 7,7,32
> > + or 7,7,9
> > + mr 3,7
> > + blr
>
> This works but we should add "r" to the register names.
>
And merge the last two instructions as a single "or r3,r7,r9".
Gabriel
^ permalink raw reply [flat|nested] 18+ messages in thread* [PATCH v3] powerpc: provide __bswapdi2
2013-05-13 7:33 ` [PATCH] " Gabriel Paubert
@ 2013-05-13 10:23 ` Michael Neuling
0 siblings, 0 replies; 18+ messages in thread
From: Michael Neuling @ 2013-05-13 10:23 UTC (permalink / raw)
To: Benjamin Herrenschmidt
Cc: mikey, David Woodhouse, linuxppc-dev, viro, David Woodhouse
From: David Woodhouse <David.Woodhouse@intel.com>
Some versions of GCC apparently expect this to be provided by libgcc.
Updates from Mikey to fix 32 bit version and adding "r" to registers.
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Michael Neuling <mikey@neuling.org>
---
Fixes from Gabriel.
arch/powerpc/kernel/misc_32.S | 11 +++++++++++
arch/powerpc/kernel/misc_64.S | 11 +++++++++++
arch/powerpc/kernel/ppc_ksyms.c | 3 ++-
3 files changed, 24 insertions(+), 1 deletion(-)
diff --git a/arch/powerpc/kernel/misc_32.S b/arch/powerpc/kernel/misc_32.S
index 19e096b..e469f30 100644
--- a/arch/powerpc/kernel/misc_32.S
+++ b/arch/powerpc/kernel/misc_32.S
@@ -657,6 +657,17 @@ _GLOBAL(__ucmpdi2)
li r3,2
blr
+_GLOBAL(__bswapdi2)
+ rotlwi r9,r4,8
+ rotlwi r10,r3,8
+ rlwimi r9,r4,24,0,7
+ rlwimi r10,r3,24,0,7
+ rlwimi r9,r4,24,16,23
+ rlwimi r10,r3,24,16,23
+ mr r3,r9
+ mr r4,r10
+ blr
+
_GLOBAL(abs)
srawi r4,r3,31
xor r3,r3,r4
diff --git a/arch/powerpc/kernel/misc_64.S b/arch/powerpc/kernel/misc_64.S
index 5cfa800..6820e45 100644
--- a/arch/powerpc/kernel/misc_64.S
+++ b/arch/powerpc/kernel/misc_64.S
@@ -234,6 +234,17 @@ _GLOBAL(__flush_dcache_icache)
isync
blr
+_GLOBAL(__bswapdi2)
+ srdi r8,r3,32
+ rlwinm r7,r3,8,0xffffffff
+ rlwimi r7,r3,24,0,7
+ rlwinm r9,r8,8,0xffffffff
+ rlwimi r7,r3,24,16,23
+ rlwimi r9,r8,24,0,7
+ rlwimi r9,r8,24,16,23
+ sldi r7,r7,32
+ or r3,r7,r9
+ blr
#if defined(CONFIG_PPC_PMAC) || defined(CONFIG_PPC_MAPLE)
/*
diff --git a/arch/powerpc/kernel/ppc_ksyms.c b/arch/powerpc/kernel/ppc_ksyms.c
index 78b8766..c296665 100644
--- a/arch/powerpc/kernel/ppc_ksyms.c
+++ b/arch/powerpc/kernel/ppc_ksyms.c
@@ -143,7 +143,8 @@ EXPORT_SYMBOL(__lshrdi3);
int __ucmpdi2(unsigned long long, unsigned long long);
EXPORT_SYMBOL(__ucmpdi2);
#endif
-
+long long __bswapdi2(long long);
+EXPORT_SYMBOL(__bswapdi2);
EXPORT_SYMBOL(memcpy);
EXPORT_SYMBOL(memset);
EXPORT_SYMBOL(memmove);
--
1.7.10.4
^ permalink raw reply related [flat|nested] 18+ messages in thread
* Re: [PATCH] powerpc: provide __bswapdi2
2013-05-13 7:09 ` Michael Neuling
2013-05-13 7:20 ` [PATCH v2] " Michael Neuling
2013-05-13 7:33 ` [PATCH] " Gabriel Paubert
@ 2013-05-13 10:33 ` David Woodhouse
2013-05-13 10:38 ` David Woodhouse
2013-05-13 16:55 ` Segher Boessenkool
2013-05-14 1:09 ` Stephen Rothwell
3 siblings, 2 replies; 18+ messages in thread
From: David Woodhouse @ 2013-05-13 10:33 UTC (permalink / raw)
To: Michael Neuling; +Cc: linuxppc-dev, viro
[-- Attachment #1: Type: text/plain, Size: 1092 bytes --]
On Mon, 2013-05-13 at 17:09 +1000, Michael Neuling wrote:
>
> This doesn't work for me but the below does:
...
>
> stolen from GCC -02 output of:
> unsigned long long __bswapdi2(unsigned long long x)
Hm, so was mine:
[dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32
.file "bswapdi2.c"
.section ".text"
.align 2
.p2align 4,,15
.globl __bswapdi2
.type __bswapdi2, @function
__bswapdi2:
rlwinm 10,4,8,0xffffffff
rlwinm 11,3,8,0xffffffff
rlwimi 10,4,24,0,7
rlwimi 11,3,24,0,7
rlwimi 10,4,24,16,23
rlwimi 11,3,24,16,23
mr 4,11
mr 3,10
blr
.size __bswapdi2,.-__bswapdi2
.ident "GCC: (GNU) 4.7.2 20121105 (Red Hat 4.7.2-2.aa.20121114svn)"
On Mon, 2013-05-13 at 09:33 +0200, Gabriel Paubert wrote:
> Actually, I'd swap the two mr instructions to never
> have an instruction that uses the result from the
> previous one.
Bad GCC. No biscuit.
Should we file a PR?
--
David Woodhouse Open Source Technology Centre
David.Woodhouse@intel.com Intel Corporation
[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 5745 bytes --]
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH] powerpc: provide __bswapdi2
2013-05-13 10:33 ` [PATCH] " David Woodhouse
@ 2013-05-13 10:38 ` David Woodhouse
2013-05-13 10:51 ` Joakim Tjernlund
2013-05-13 13:12 ` Gabriel Paubert
2013-05-13 16:55 ` Segher Boessenkool
1 sibling, 2 replies; 18+ messages in thread
From: David Woodhouse @ 2013-05-13 10:38 UTC (permalink / raw)
To: Michael Neuling; +Cc: linuxppc-dev, viro
[-- Attachment #1: Type: text/plain, Size: 1308 bytes --]
On Mon, 2013-05-13 at 11:33 +0100, David Woodhouse wrote:
>
> On Mon, 2013-05-13 at 09:33 +0200, Gabriel Paubert wrote:
> > Actually, I'd swap the two mr instructions to never
> > have an instruction that uses the result from the
> > previous one.
>
> Bad GCC. No biscuit.
>
> Should we file a PR?
Maybe not. If you tell it to tune for an in-order machine like Cell, it
swaps them round. Although now I'm confused about which of POWER[567]
were in-order:
[dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 | grep -B1 mr
rlwimi 11,3,24,16,23
mr 4,11
mr 3,10
[dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 -mtune=cell | grep -B1 mr
rlwimi 11,3,24,16,23
mr 3,10
mr 4,11
[dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 -mtune=power5 | grep -B1 mr
rlwimi 11,3,24,16,23
mr 3,10
mr 4,11
[dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 -mtune=power6 | grep -B1 mr
rlwimi 11,3,24,16,23
mr 4,11
mr 3,10
[dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 -mtune=power7 | grep -B1 mr
rlwimi 11,3,24,16,23
mr 4,11
mr 3,10
--
David Woodhouse Open Source Technology Centre
David.Woodhouse@intel.com Intel Corporation
[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 5745 bytes --]
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH] powerpc: provide __bswapdi2
2013-05-13 10:38 ` David Woodhouse
@ 2013-05-13 10:51 ` Joakim Tjernlund
2013-05-13 11:12 ` Joakim Tjernlund
2013-05-13 13:12 ` Gabriel Paubert
1 sibling, 1 reply; 18+ messages in thread
From: Joakim Tjernlund @ 2013-05-13 10:51 UTC (permalink / raw)
To: David Woodhouse; +Cc: linuxppc-dev, Michael Neuling, viro
"Linuxppc-dev"
<linuxppc-dev-bounces+joakim.tjernlund=transmode.se@lists.ozlabs.org>
wrote 2013/05/13 12:38:13:
>
> On Mon, 2013-05-13 at 11:33 +0100, David Woodhouse wrote:
> >
> > On Mon, 2013-05-13 at 09:33 +0200, Gabriel Paubert wrote:
> > > Actually, I'd swap the two mr instructions to never
> > > have an instruction that uses the result from the
> > > previous one.
> >
> > Bad GCC. No biscuit.
> >
> > Should we file a PR?
>
> Maybe not. If you tell it to tune for an in-order machine like Cell, it
> swaps them round. Although now I'm confused about which of POWER[567]
> were in-order:
>
> [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 | grep
-B1 mr
> rlwimi 11,3,24,16,23
> mr 4,11
> mr 3,10
> [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32
-mtune=cell | grep -B1 mr
> rlwimi 11,3,24,16,23
> mr 3,10
> mr 4,11
> [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32
-mtune=power5 | grep -B1 mr
> rlwimi 11,3,24,16,23
> mr 3,10
> mr 4,11
> [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32
-mtune=power6 | grep -B1 mr
> rlwimi 11,3,24,16,23
> mr 4,11
> mr 3,10
> [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32
-mtune=power7 | grep -B1 mr
> rlwimi 11,3,24,16,23
> mr 4,11
> mr 3,10
A bit rusty on the ppc asm but can you not remove the mr completely:
rlwimi 10,4,24,16,23
rlwimi 11,3,24,16,23
mr 4,11
mr 3,10
to
rlwimi 4,4,24,16,23
rlwimi 3,3,24,16,23
Jocke
^ permalink raw reply [flat|nested] 18+ messages in thread* Re: [PATCH] powerpc: provide __bswapdi2
2013-05-13 10:51 ` Joakim Tjernlund
@ 2013-05-13 11:12 ` Joakim Tjernlund
0 siblings, 0 replies; 18+ messages in thread
From: Joakim Tjernlund @ 2013-05-13 11:12 UTC (permalink / raw)
Cc: linuxppc-dev, Michael Neuling, David Woodhouse, viro
"Linuxppc-dev"
<linuxppc-dev-bounces+joakim.tjernlund=transmode.se@lists.ozlabs.org>
wrote on 2013/05/13 12:51:59:
>
> "Linuxppc-dev"
> <linuxppc-dev-bounces+joakim.tjernlund=transmode.se@lists.ozlabs.org>
> wrote 2013/05/13 12:38:13:
> >
> > On Mon, 2013-05-13 at 11:33 +0100, David Woodhouse wrote:
> > >
> > > On Mon, 2013-05-13 at 09:33 +0200, Gabriel Paubert wrote:
> > > > Actually, I'd swap the two mr instructions to never
> > > > have an instruction that uses the result from the
> > > > previous one.
> > >
> > > Bad GCC. No biscuit.
> > >
> > > Should we file a PR?
> >
> > Maybe not. If you tell it to tune for an in-order machine like Cell,
it
> > swaps them round. Although now I'm confused about which of POWER[567]
> > were in-order:
> >
> > [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 |
grep
> -B1 mr
> > rlwimi 11,3,24,16,23
> > mr 4,11
> > mr 3,10
> > [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32
> -mtune=cell | grep -B1 mr
> > rlwimi 11,3,24,16,23
> > mr 3,10
> > mr 4,11
> > [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32
> -mtune=power5 | grep -B1 mr
> > rlwimi 11,3,24,16,23
> > mr 3,10
> > mr 4,11
> > [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32
> -mtune=power6 | grep -B1 mr
> > rlwimi 11,3,24,16,23
> > mr 4,11
> > mr 3,10
> > [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32
> -mtune=power7 | grep -B1 mr
> > rlwimi 11,3,24,16,23
> > mr 4,11
> > mr 3,10
>
> A bit rusty on the ppc asm but can you not remove the mr completely:
> rlwimi 10,4,24,16,23
> rlwimi 11,3,24,16,23
> mr 4,11
> mr 3,10
> to
> rlwimi 4,4,24,16,23
> rlwimi 3,3,24,16,23
Oops, that got twisted. Forget my comment.
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH] powerpc: provide __bswapdi2
2013-05-13 10:38 ` David Woodhouse
2013-05-13 10:51 ` Joakim Tjernlund
@ 2013-05-13 13:12 ` Gabriel Paubert
1 sibling, 0 replies; 18+ messages in thread
From: Gabriel Paubert @ 2013-05-13 13:12 UTC (permalink / raw)
To: David Woodhouse; +Cc: linuxppc-dev, Michael Neuling, viro
On Mon, May 13, 2013 at 11:38:13AM +0100, David Woodhouse wrote:
> On Mon, 2013-05-13 at 11:33 +0100, David Woodhouse wrote:
> >
> > On Mon, 2013-05-13 at 09:33 +0200, Gabriel Paubert wrote:
> > > Actually, I'd swap the two mr instructions to never
> > > have an instruction that uses the result from the
> > > previous one.
> >
> > Bad GCC. No biscuit.
> >
> > Should we file a PR?
>
> Maybe not. If you tell it to tune for an in-order machine like Cell, it
> swaps them round. Although now I'm confused about which of POWER[567]
> were in-order:
It was Power6 IIRC. On this kind of fine point, don't rely too much
on what GCC produces.
>
> [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 | grep -B1 mr
> rlwimi 11,3,24,16,23
> mr 4,11
> mr 3,10
> [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 -mtune=cell | grep -B1 mr
> rlwimi 11,3,24,16,23
> mr 3,10
> mr 4,11
> [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 -mtune=power5 | grep -B1 mr
> rlwimi 11,3,24,16,23
> mr 3,10
> mr 4,11
> [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 -mtune=power6 | grep -B1 mr
> rlwimi 11,3,24,16,23
> mr 4,11
> mr 3,10
> [dwmw2@i7 ~]$ powerpc64-linux-gnu-gcc -O2 -S -o- bswapdi2.c -m32 -mtune=power7 | grep -B1 mr
> rlwimi 11,3,24,16,23
> mr 4,11
> mr 3,10
I don't know of any processor in which putting the mr 3,10 first can cause stalls, so
even a generic tuning should put it first.
Gabriel
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH] powerpc: provide __bswapdi2
2013-05-13 10:33 ` [PATCH] " David Woodhouse
2013-05-13 10:38 ` David Woodhouse
@ 2013-05-13 16:55 ` Segher Boessenkool
1 sibling, 0 replies; 18+ messages in thread
From: Segher Boessenkool @ 2013-05-13 16:55 UTC (permalink / raw)
To: David Woodhouse; +Cc: linuxppc-dev, Michael Neuling, viro
> rlwinm 10,4,8,0xffffffff
> rlwinm 11,3,8,0xffffffff
> rlwimi 10,4,24,0,7
> rlwimi 11,3,24,0,7
> rlwimi 10,4,24,16,23
> rlwimi 11,3,24,16,23
> mr 4,11
> mr 3,10
>> Actually, I'd swap the two mr instructions to never
>> have an instruction that uses the result from the
>> previous one.
>
> Bad GCC. No biscuit.
>
> Should we file a PR?
This is scheduled just fine. Every pair of instructions here can
execute together (on most CPUs, if not all); all instructions after
it are dependent on previous instructions. There also is no issue
(group) restriction that makes this scheduling suboptimal afaics.
Segher
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH] powerpc: provide __bswapdi2
2013-05-13 7:09 ` Michael Neuling
` (2 preceding siblings ...)
2013-05-13 10:33 ` [PATCH] " David Woodhouse
@ 2013-05-14 1:09 ` Stephen Rothwell
2013-05-14 1:25 ` Michael Neuling
3 siblings, 1 reply; 18+ messages in thread
From: Stephen Rothwell @ 2013-05-14 1:09 UTC (permalink / raw)
To: Michael Neuling; +Cc: linuxppc-dev, David Woodhouse, viro
[-- Attachment #1: Type: text/plain, Size: 1000 bytes --]
Hi Mikey,
On Mon, 13 May 2013 17:09:59 +1000 Michael Neuling <mikey@neuling.org> wrote:
>
> This doesn't work for me but the below does:
>
> _GLOBAL(__bswapdi2)
> rotlwi r9,r4,8
> rotlwi r10,r3,8
> rlwimi r9,r4,24,0,7
> rlwimi r10,r3,24,0,7
> rlwimi r9,r4,24,16,23
> rlwimi r10,r3,24,16,23
> mr r4,r10
> mr r3,r9
> blr
>
> stolen from GCC -02 output of:
> unsigned long long __bswapdi2(unsigned long long x)
> {
> return ((x & 0x00000000000000ffULL) << 56) |
> ((x & 0x000000000000ff00ULL) << 40) |
> ((x & 0x0000000000ff0000ULL) << 24) |
> ((x & 0x00000000ff000000ULL) << 8) |
> ((x & 0x000000ff00000000ULL) >> 8) |
> ((x & 0x0000ff0000000000ULL) >> 24) |
> ((x & 0x00ff000000000000ULL) >> 40) |
> ((x & 0xff00000000000000ULL) >> 56);
> }
So, if we are just stealing the output of gcc, why not just use the C
version (at least for 32 bit)?
--
Cheers,
Stephen Rothwell sfr@canb.auug.org.au
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 18+ messages in thread* Re: [PATCH] powerpc: provide __bswapdi2
2013-05-14 1:09 ` Stephen Rothwell
@ 2013-05-14 1:25 ` Michael Neuling
2013-05-14 6:59 ` David Woodhouse
0 siblings, 1 reply; 18+ messages in thread
From: Michael Neuling @ 2013-05-14 1:25 UTC (permalink / raw)
To: Stephen Rothwell; +Cc: Linux PPC dev, David Woodhouse, viro
[-- Attachment #1: Type: text/plain, Size: 1201 bytes --]
On 14 May 2013 11:09, "Stephen Rothwell" <sfr@canb.auug.org.au> wrote:
>
> Hi Mikey,
>
> On Mon, 13 May 2013 17:09:59 +1000 Michael Neuling <mikey@neuling.org>
wrote:
> >
> > This doesn't work for me but the below does:
> >
> > _GLOBAL(__bswapdi2)
> > rotlwi r9,r4,8
> > rotlwi r10,r3,8
> > rlwimi r9,r4,24,0,7
> > rlwimi r10,r3,24,0,7
> > rlwimi r9,r4,24,16,23
> > rlwimi r10,r3,24,16,23
> > mr r4,r10
> > mr r3,r9
> > blr
> >
> > stolen from GCC -02 output of:
> > unsigned long long __bswapdi2(unsigned long long x)
> > {
> > return ((x & 0x00000000000000ffULL) << 56) |
> > ((x & 0x000000000000ff00ULL) << 40) |
> > ((x & 0x0000000000ff0000ULL) << 24) |
> > ((x & 0x00000000ff000000ULL) << 8) |
> > ((x & 0x000000ff00000000ULL) >> 8) |
> > ((x & 0x0000ff0000000000ULL) >> 24) |
> > ((x & 0x00ff000000000000ULL) >> 40) |
> > ((x & 0xff00000000000000ULL) >> 56);
> > }
>
> So, if we are just stealing the output of gcc, why not just use the C
> version (at least for 32 bit)?
Woodhouse: can we just do this?
Mikey
[-- Attachment #2: Type: text/html, Size: 1793 bytes --]
^ permalink raw reply [flat|nested] 18+ messages in thread* Re: [PATCH] powerpc: provide __bswapdi2
2013-05-14 1:25 ` Michael Neuling
@ 2013-05-14 6:59 ` David Woodhouse
0 siblings, 0 replies; 18+ messages in thread
From: David Woodhouse @ 2013-05-14 6:59 UTC (permalink / raw)
To: Michael Neuling; +Cc: Stephen Rothwell, viro, Linux PPC dev
[-- Attachment #1: Type: text/plain, Size: 1235 bytes --]
On Tue, 2013-05-14 at 11:25 +1000, Michael Neuling wrote:
>
> > So, if we are just stealing the output of gcc, why not just use the C
> > version (at least for 32 bit)?
>
> Woodhouse: can we just do this?
Sure, if you don't mind GCC optimising the contents of your C function
by turning it into a call to libgcc's __bswapdi2() :)
OK, you might be able to do some archaeology and determine that the only
compiler that emits calls to __bswapdi2() is GCC 4.4, and furthermore
that the same compiler *doesn't* have the wit to notice that the
contents of the function are a 64-bit byteswap, so it's never going to
happen. But I don't like that approach. I'd feel I have to sacrifice a
goat *anyway*, and I don't have a spare goat.
Although now I come to explicitly explain why I did it that way... it
occurs to me that the libgcc version is just written in C, and the
compiler evidently trusts itself not to optimise that into a recursive
call. Is there a compiler switch which guarantees that, which we could
use without other unwanted side-effects?
--
David Woodhouse Open Source Technology Centre
David.Woodhouse@intel.com Intel Corporation
[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 5745 bytes --]
^ permalink raw reply [flat|nested] 18+ messages in thread