On Jun 18, 2015 3:41 AM, "Borislav Petkov" <bp@xxxxxxx> wrote:
>
> On Thu, Jun 18, 2015 at 03:59:06AM -0400, George Spelvin wrote:
> > Before, the code to do rdtsc looked like:
> > rdtsc
> > shl $0x20,%rdx
> > mov %eax,%eax
> > or %rdx,%rax
> >
> > The "mov %eax,%eax" is required to clear the high 32 bits of %rax.
> >
> > By declaring low and high as 64-bit variables, the code is simplified to
> > rdtsc
> > shl $0x20,%rdx
> > or %rdx,%rax
> >
> > Yes, it's a 2-byte instruction that's not on a critical path, but
> > there are principles to be upheld.
> >
> > Every user of EAX_EDX_RET has been checked. I tried to check users of
> > EAX_EDX_ARGS, but there weren't any, so I deleted it to be safe.
> >
> > (There's no benefit to making "high" 64 bits, but it was the simplest
> > way to proceed.)
> >
> > Signed-off-by: George Spelvin <linux@xxxxxxxxxxx>
> > ---
> > I came across this while in the middle of hacking on some other code that
> > isn't booting yet, so I haven't exactly tested it thoroughly. I *did*
> > check both the Intel and AMD specifications to ensure that this was a safe
> > optimization, and did user-space tests on both Intel and AMD processors.
> >
> > But it's simple enough I'm willing to risk public embarrassment.
> >
> > arch/x86/include/asm/msr.h | 7 +++----
> > 1 file changed, 3 insertions(+), 4 deletions(-)
> >
> > diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
> > index de36f22e..d6b8a80d 100644
> > --- a/arch/x86/include/asm/msr.h
> > +++ b/arch/x86/include/asm/msr.h
> > @@ -46,14 +46,13 @@ static inline unsigned long long native_read_tscp(unsigned int *aux)
> > * it means rax *or* rdx.
> > */
> > #ifdef CONFIG_X86_64
> > -#define DECLARE_ARGS(val, low, high) unsigned low, high
> > -#define EAX_EDX_VAL(val, low, high) ((low) | ((u64)(high) << 32))
> > -#define EAX_EDX_ARGS(val, low, high) "a" (low), "d" (high)
> > +/* Using 64-bit values saves one instruction clearing the high half of low */
> > +#define DECLARE_ARGS(val, low, high) unsigned long low, high
> > +#define EAX_EDX_VAL(val, low, high) ((low) | (high) << 32)
> > #define EAX_EDX_RET(val, low, high) "=a" (low), "=d" (high)
> > #else
> > #define DECLARE_ARGS(val, low, high) unsigned long long val
> > #define EAX_EDX_VAL(val, low, high) (val)
> > -#define EAX_EDX_ARGS(val, low, high) "A" (val)
> > #define EAX_EDX_RET(val, low, high) "=A" (val)
> > #endif
>
> Looks ok to me, I'll pick it up along with Andy's TSC cleanup. CCed.
>
Weird. Why does gcc think that an "=a" constraint on a 32-bit
variable results in undefined high bits?
I just compiled this:
unsigned long long merge(unsigned int low, unsigned int high)
{
return (low) | ((unsigned long long)(high) << 32);
}
and I get this:
merge:
movq %rsi, %rax
movl %edi, %edi
salq $32, %rax
orq %rdi, %rax
ret
Huh? Clang does the same thing.
In fact, if I compile:
extern unsigned int func(void);
unsigned long long promote(void)
{
return func();
}
I get:
promote:
subq $8, %rsp
call func
addq $8, %rsp
movl %eax, %eax
ret
Now I'm extra confused. Do I just misunderstand the ABI, or are both
clang and gcc missing an obvious optimization?
--Andy
--
To unsubscribe from this list: send the line "unsubscribe linux-x86_64" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
