RE: [PATCH v5 2/2] riscv: Optimize bitops with Zbb extension

[Wang, Xiao W]

Hi Geert,

> -----Original Message-----
> From: Geert Uytterhoeven <geert@xxxxxxxxxxxxxx>
> Sent: Friday, November 10, 2023 5:25 PM
> To: Wang, Xiao W <xiao.w.wang@xxxxxxxxx>
> Cc: paul.walmsley@xxxxxxxxxx; palmer@xxxxxxxxxxx;
> aou@xxxxxxxxxxxxxxxxx; ardb@xxxxxxxxxx; anup@xxxxxxxxxxxxxx; Li, Haicheng
> <haicheng.li@xxxxxxxxx>; ajones@xxxxxxxxxxxxxxxx; Liu, Yujie
> <yujie.liu@xxxxxxxxx>; charlie@xxxxxxxxxxxx; linux-riscv@xxxxxxxxxxxxxxxxxxx;
> linux-efi@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> Subject: Re: [PATCH v5 2/2] riscv: Optimize bitops with Zbb extension
> 
> Hi Xiao,
> 
> On Tue, Oct 31, 2023 at 7:37 AM Xiao Wang <xiao.w.wang@xxxxxxxxx> wrote:
> > This patch leverages the alternative mechanism to dynamically optimize
> > bitops (including __ffs, __fls, ffs, fls) with Zbb instructions. When
> > Zbb ext is not supported by the runtime CPU, legacy implementation is
> > used. If Zbb is supported, then the optimized variants will be selected
> > via alternative patching.
> >
> > The legacy bitops support is taken from the generic C implementation as
> > fallback.
> >
> > If the parameter is a build-time constant, we leverage compiler builtin to
> > calculate the result directly, this approach is inspired by x86 bitops
> > implementation.
> >
> > EFI stub runs before the kernel, so alternative mechanism should not be
> > used there, this patch introduces a macro NO_ALTERNATIVE for this
> purpose.
> >
> > Signed-off-by: Xiao Wang <xiao.w.wang@xxxxxxxxx>
> > Reviewed-by: Charlie Jenkins <charlie@xxxxxxxxxxxx>
> 
> Thanks for your patch, which is now commit 457926b253200bd9 ("riscv:
> Optimize bitops with Zbb extension") in riscv/for-next.
> 
> > --- a/arch/riscv/include/asm/bitops.h
> > +++ b/arch/riscv/include/asm/bitops.h
> > @@ -15,13 +15,261 @@
> >  #include <asm/barrier.h>
> >  #include <asm/bitsperlong.h>
> >
> > +#if !defined(CONFIG_RISCV_ISA_ZBB) || defined(NO_ALTERNATIVE)
> >  #include <asm-generic/bitops/__ffs.h>
> > -#include <asm-generic/bitops/ffz.h>
> > -#include <asm-generic/bitops/fls.h>
> >  #include <asm-generic/bitops/__fls.h>
> > +#include <asm-generic/bitops/ffs.h>
> > +#include <asm-generic/bitops/fls.h>
> > +
> > +#else
> > +#include <asm/alternative-macros.h>
> > +#include <asm/hwcap.h>
> > +
> > +#if (BITS_PER_LONG == 64)
> > +#define CTZW   "ctzw "
> > +#define CLZW   "clzw "
> > +#elif (BITS_PER_LONG == 32)
> > +#define CTZW   "ctz "
> > +#define CLZW   "clz "
> > +#else
> > +#error "Unexpected BITS_PER_LONG"
> > +#endif
> > +
> > +static __always_inline unsigned long variable__ffs(unsigned long word)
> > +{
> > +       int num;
> > +
> > +       asm_volatile_goto(ALTERNATIVE("j %l[legacy]", "nop", 0,
> > +                                     RISCV_ISA_EXT_ZBB, 1)
> > +                         : : : : legacy);
> > +
> > +       asm volatile (".option push\n"
> > +                     ".option arch,+zbb\n"
> > +                     "ctz %0, %1\n"
> > +                     ".option pop\n"
> > +                     : "=r" (word) : "r" (word) :);
> > +
> > +       return word;
> > +
> > +legacy:
> > +       num = 0;
> > +#if BITS_PER_LONG == 64
> > +       if ((word & 0xffffffff) == 0) {
> > +               num += 32;
> > +               word >>= 32;
> > +       }
> > +#endif
> > +       if ((word & 0xffff) == 0) {
> > +               num += 16;
> > +               word >>= 16;
> > +       }
> > +       if ((word & 0xff) == 0) {
> > +               num += 8;
> > +               word >>= 8;
> > +       }
> > +       if ((word & 0xf) == 0) {
> > +               num += 4;
> > +               word >>= 4;
> > +       }
> > +       if ((word & 0x3) == 0) {
> > +               num += 2;
> > +               word >>= 2;
> > +       }
> > +       if ((word & 0x1) == 0)
> > +               num += 1;
> > +       return num;
> > +}
> 
> Surely we can do better than duplicating include/asm-generic/bitops/__ffs.h?
> 
> E.g. rename the generic implementation to generic___ffs():
> 
>     -static __always_inline unsigned long __ffs(unsigned long word)
>     +static __always_inline unsigned long generic__ffs(unsigned long word)
>      {
>              ...
>      }
> 
>     +#ifndef __ffs
>     +#define __ffs(x) generic__ffs(x)
>     +#endif
> 
> and explicitly calling the generic one here?
> 
> Same comment for the other functions.

Thanks for the suggestion. I just tried your above example and got build error of
"__ffs" redefinition, I think we can change the macro condition to:

+#ifndef __HAVE_ARCH___FFS
+#define __ffs(word) generic___ffs(word)
+#endif

Besides, adding a "generic_" prefix to __ffs would make it as "generic___ffs".
I saw similar API names in include/asm-generic/bitops/generic-non-atomic.h.

I would send a patch for this.

BRs,
Xiao

> 
> Gr{oetje,eeting}s,
> 
>                         Geert
> 
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-
> m68k.org
> 
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                 -- Linus Torvalds