Re: need a fast byteorder changing routine

Re: need a fast byteorder changing routine

[Wojciech Kromer]

Użytkownik li napisał:

>Hi all,
>i need do a lot of byteorder changing operation from big-endian to little-endian. cpu_to_le32 is not fast enough for me, does anyone have such a routine written in asm.?
>PS: here is an ugly asm function do such thing:
>void swapbyte(unsigned long *val, unsigned long offset,
>              unsigned long value)
>{
>
>        asm ("lis 3 %0@ha" : "=r" (val));
>        asm ("lwz 3,%0@l(3)" : "=r" (val));
>        asm ("stw %1, 0(4)" : "=r" (offset));
>        asm ("stw %2, 0(5)" : "=r" (value));
>        asm ("stwbrx 5, 3, 4");
>}val is the address of value. i used to set this function as inline, but kernel said:
>Oops: kernel access of bad area, sig: 11
>Any comments
>Best Regards!
>Li
>
>
>
>
>
try DIRECT using out_le16/32 in_le16/32 from <asm/io.h>,
this takes same time as one  big endian write/read



--
* * * * * * * * * * * *
* per pedes ad astra! *
* * * * * * * * * * * *    mailto:krom@dgt-lab.com.pl


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: need a fast byteorder changing routine

[Peter Barada]

>Li,
>
>If you must do it one word at a time, the following works.
>
>long SwapBytes(long* p)
>{
>  __asm__(" lwbrx 3,0,3");
>}
>
>Charlie

And if you want it inlined:

static inline unsigned long loadWordAndSwapBytes(unsigned long *p)
{
  unsigned long val;

  __asm__ ("lwbrx %0,0,%1" : "=r" (val) : "r" (p));
  return val;
}

--
Peter Barada                             Peter.Barada@motorola.com
Wizard                                   781-852-2768 (direct)
WaveMark Solutions(A Motorola Company)   781-270-0193 (fax)

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

RE: need a fast byteorder changing routine

[Wells, Charles]

Li,

If you must do it one word at a time, the following works.

long SwapBytes(long* p)
{
  __asm__(" lwbrx 3,0,3");
}

Charlie

-----Original Message-----
From: Wells, Charles [mailto:Charles.Wells@nielsenmedia.com]
Sent: Friday, July 18, 2003 9:04 AM
To: 'li'
Cc: linuxppc-embedded@lists.linuxppc.org
Subject: RE: need a fast byteorder changing routine



Li,

There's several ways of doing it.  This way is pretty fast and makes few
assumptions about the buffer.

void FixEndian(long* buf, unsigned long num_longs)
{
  __asm__(" addi   3,3,-4");
  __asm__(" mtspr  9,4");
  __asm__(" lwzu   4,4(3)");
  __asm__(" stwbrx 4,0,3");
  __asm__(" bc     16,0,$-8");
}

Charlie


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

RE: need a fast byteorder changing routine

[Wells, Charles]

Li,

There's several ways of doing it.  This way is pretty fast and makes few
assumptions about the buffer.

void FixEndian(long* buf, unsigned long num_longs)
{
  __asm__(" addi   3,3,-4");
  __asm__(" mtspr  9,4");
  __asm__(" lwzu   4,4(3)");
  __asm__(" stwbrx 4,0,3");
  __asm__(" bc     16,0,$-8");
}

Charlie

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/