Re: why not put 64 bit value directly to register

[Thiemo Seufer <ths@networkno.de>]

Bin Chen wrote:
> Hi,
> 
> This code is snip from u-boot, I don't know why the 32bit-64bit conversion
> is needed, why not put val directly to register but do the transform?
> 
> static void cvmx_write_cop0_entry_lo_0(uint64_t val)
> {
>     uint32_t val_low  = val & 0xffffffff;
>     uint32_t val_high = val  >> 32;
> 
>     uint32_t tmp; /* temp register */
> 
>     asm volatile (
>         "  .set mips64                       \n"
>         "  .set noreorder                    \n"
>         /* Standard twin 32 bit -> 64 bit construction */
>         "  dsll  %[valh], 32                 \n"
>         "  dla   %[tmp], 0xffffffff          \n"
>         "  and   %[vall], %[tmp], %[vall]    \n"
>         "  daddu %[valh], %[valh], %[vall]   \n"
>         /* Combined value is in valh */
>         "  dmtc0 %[valh],$2,0                \n"
>         "  .set reorder                      \n"
>          :[tmp] "=&r" (tmp) : [valh] "r" (val_high), [vall] "r" (val_low) );
> }

This will convert on a 64bit capable MIPS CPU an o32 ABI register pair
holding a long long to a 64bit register value, and write that to a CP0
register.

It will break if an exception happens in between unless the exception
handlers save/restore the upper half as well.


Thiemo