[PATCH] x86: use tzcnt instead of bsf

[PATCH] x86: use tzcnt instead of bsf

[Jan Beulich]

[-- Attachment #1: Type: text/plain, Size: 1798 bytes --]

Following a compiler change done in 2012, make use of the fact that for
non-zero input BSF and TZCNT produce the same numeric result (EFLAGS
setting differs), and that CPUs not knowing of TZCNT will treat the
instruction as BSF (i.e. ignore what looks like a REP prefix to them).
The assumption here is that TZCNT would never have worse performance
than BSF.

Also extend the asm() input in find_first_set_bit() to allow memory
operands.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
Thanks to Andrew for noticing that I forgot to post this for Xen after
a similar change got accepted into the Linux kernel.

--- a/xen/arch/x86/bitops.c
+++ b/xen/arch/x86/bitops.c
@@ -62,7 +62,7 @@ unsigned int __find_first_zero_bit(
         "   je 2f\n\t"
         "   xor -"STR(BITS_PER_LONG/8)"(%2),%3\n\t"
         "   jz 1b\n\t"
-        "   bsf %3,%0\n\t"
+        "   rep; bsf %3,%0\n\t"
         "   lea -"STR(BITS_PER_LONG/8)"(%2),%2\n\t"
         "2: sub %%ebx,%%edi\n\t"
         "   shl $3,%%edi\n\t"
--- a/xen/arch/x86/hvm/vpic.c
+++ b/xen/arch/x86/hvm/vpic.c
@@ -56,7 +56,7 @@ static int vpic_get_priority(struct hvm_
         return VPIC_PRIO_NONE;
 
     /* prio = ffs(mask ROR vpic->priority_add); */
-    asm ( "ror %%cl,%b1 ; bsf %1,%0"
+    asm ( "ror %%cl,%b1 ; rep; bsf %1,%0"
           : "=r" (prio) : "q" ((uint32_t)mask), "c" (vpic->priority_add) );
     return prio;
 }
--- a/xen/include/asm-x86/bitops.h
+++ b/xen/include/asm-x86/bitops.h
@@ -382,7 +382,7 @@ static inline unsigned int __scanbit(uns
  */
 static inline unsigned int find_first_set_bit(unsigned long word)
 {
-    asm ( "bsf %1,%0" : "=r" (word) : "r" (word) );
+    asm ( "rep; bsf %1,%0" : "=r" (word) : "rm" (word) );
     return (unsigned int)word;
 }
 




[-- Attachment #2: x86-use-tzcnt.patch --]
[-- Type: text/plain, Size: 1825 bytes --]

x86: use tzcnt instead of bsf

Following a compiler change done in 2012, make use of the fact that for
non-zero input BSF and TZCNT produce the same numeric result (EFLAGS
setting differs), and that CPUs not knowing of TZCNT will treat the
instruction as BSF (i.e. ignore what looks like a REP prefix to them).
The assumption here is that TZCNT would never have worse performance
than BSF.

Also extend the asm() input in find_first_set_bit() to allow memory
operands.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
Thanks to Andrew for noticing that I forgot to post this for Xen after
a similar change got accepted into the Linux kernel.

--- a/xen/arch/x86/bitops.c
+++ b/xen/arch/x86/bitops.c
@@ -62,7 +62,7 @@ unsigned int __find_first_zero_bit(
         "   je 2f\n\t"
         "   xor -"STR(BITS_PER_LONG/8)"(%2),%3\n\t"
         "   jz 1b\n\t"
-        "   bsf %3,%0\n\t"
+        "   rep; bsf %3,%0\n\t"
         "   lea -"STR(BITS_PER_LONG/8)"(%2),%2\n\t"
         "2: sub %%ebx,%%edi\n\t"
         "   shl $3,%%edi\n\t"
--- a/xen/arch/x86/hvm/vpic.c
+++ b/xen/arch/x86/hvm/vpic.c
@@ -56,7 +56,7 @@ static int vpic_get_priority(struct hvm_
         return VPIC_PRIO_NONE;
 
     /* prio = ffs(mask ROR vpic->priority_add); */
-    asm ( "ror %%cl,%b1 ; bsf %1,%0"
+    asm ( "ror %%cl,%b1 ; rep; bsf %1,%0"
           : "=r" (prio) : "q" ((uint32_t)mask), "c" (vpic->priority_add) );
     return prio;
 }
--- a/xen/include/asm-x86/bitops.h
+++ b/xen/include/asm-x86/bitops.h
@@ -382,7 +382,7 @@ static inline unsigned int __scanbit(uns
  */
 static inline unsigned int find_first_set_bit(unsigned long word)
 {
-    asm ( "bsf %1,%0" : "=r" (word) : "r" (word) );
+    asm ( "rep; bsf %1,%0" : "=r" (word) : "rm" (word) );
     return (unsigned int)word;
 }
 

[-- Attachment #3: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: [PATCH] x86: use tzcnt instead of bsf

[Andrew Cooper]

On 23/01/15 16:39, Jan Beulich wrote:
> Following a compiler change done in 2012, make use of the fact that for
> non-zero input BSF and TZCNT produce the same numeric result (EFLAGS
> setting differs), and that CPUs not knowing of TZCNT will treat the
> instruction as BSF (i.e. ignore what looks like a REP prefix to them).
> The assumption here is that TZCNT would never have worse performance
> than BSF.
>
> Also extend the asm() input in find_first_set_bit() to allow memory
> operands.
>
> Signed-off-by: Jan Beulich <jbeulich@suse.com>

Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>

> ---
> Thanks to Andrew for noticing that I forgot to post this for Xen after
> a similar change got accepted into the Linux kernel.
>
> --- a/xen/arch/x86/bitops.c
> +++ b/xen/arch/x86/bitops.c
> @@ -62,7 +62,7 @@ unsigned int __find_first_zero_bit(
>          "   je 2f\n\t"
>          "   xor -"STR(BITS_PER_LONG/8)"(%2),%3\n\t"
>          "   jz 1b\n\t"
> -        "   bsf %3,%0\n\t"
> +        "   rep; bsf %3,%0\n\t"
>          "   lea -"STR(BITS_PER_LONG/8)"(%2),%2\n\t"
>          "2: sub %%ebx,%%edi\n\t"
>          "   shl $3,%%edi\n\t"
> --- a/xen/arch/x86/hvm/vpic.c
> +++ b/xen/arch/x86/hvm/vpic.c
> @@ -56,7 +56,7 @@ static int vpic_get_priority(struct hvm_
>          return VPIC_PRIO_NONE;
>  
>      /* prio = ffs(mask ROR vpic->priority_add); */
> -    asm ( "ror %%cl,%b1 ; bsf %1,%0"
> +    asm ( "ror %%cl,%b1 ; rep; bsf %1,%0"
>            : "=r" (prio) : "q" ((uint32_t)mask), "c" (vpic->priority_add) );
>      return prio;
>  }
> --- a/xen/include/asm-x86/bitops.h
> +++ b/xen/include/asm-x86/bitops.h
> @@ -382,7 +382,7 @@ static inline unsigned int __scanbit(uns
>   */
>  static inline unsigned int find_first_set_bit(unsigned long word)
>  {
> -    asm ( "bsf %1,%0" : "=r" (word) : "r" (word) );
> +    asm ( "rep; bsf %1,%0" : "=r" (word) : "rm" (word) );
>      return (unsigned int)word;
>  }
>  
>
>
>