diff mbox series

[v2,1/2] x86/asm/bitops: Replace __fls() by its generic builtin implementation

Message ID 20221125023330.195581-2-mailhol.vincent@wanadoo.fr
State New
Headers
Series x86/asm/bitops: optimize fls functions for constant expressions |

Commit Message

Vincent Mailhol Nov. 25, 2022, 2:33 a.m. UTC 
  Below snippet:

  #include <linux/bitops.h>

  unsigned int foo(unsigned long word)
  {
  	return __fls(word);
  }

produces this on GCC 12.1.0:

  0000000000000000 <foo>:
     0:	f3 0f 1e fa          	endbr64
     4:	e8 00 00 00 00       	call   9 <foo+0x9>
     9:	53                   	push   %rbx
     a:	48 89 fb             	mov    %rdi,%rbx
     d:	e8 00 00 00 00       	call   12 <foo+0x12>
    12:	48 0f bd c3          	bsr    %rbx,%rax
    16:	5b                   	pop    %rbx
    17:	31 ff                	xor    %edi,%edi
    19:	e9 00 00 00 00       	jmp    1e <foo+0x1e>

and that on clang 14.0.6:

  0000000000000000 <foo>:
     0:	f3 0f 1e fa          	endbr64
     4:	e8 00 00 00 00       	call   9 <foo+0x9>
     9:	53                   	push   %rbx
     a:	50                   	push   %rax
     b:	48 89 fb             	mov    %rdi,%rbx
     e:	e8 00 00 00 00       	call   13 <foo+0x13>
    13:	48 89 1c 24          	mov    %rbx,(%rsp)
    17:	48 0f bd 04 24       	bsr    (%rsp),%rax
    1c:	48 83 c4 08          	add    $0x8,%rsp
    20:	5b                   	pop    %rbx
    21:	c3                   	ret

The implementation from <asm-generic/bitops/builtin-__fls.h> [1]
produces the exact same code on GCC and below code on clang:

  0000000000000000 <foo>:
     0:	f3 0f 1e fa          	endbr64
     4:	e8 00 00 00 00       	call   9 <foo+0x9>
     9:	53                   	push   %rbx
     a:	48 89 fb             	mov    %rdi,%rbx
     d:	e8 00 00 00 00       	call   12 <foo+0x12>
    12:	48 0f bd c3          	bsr    %rbx,%rax
    16:	5b                   	pop    %rbx
    17:	c3                   	ret

The builtin implementation is better for two reasons:

  1/ it saves two instructions on clang (a push and a stack pointer
     decrement).

  2/ when used on constant expressions, the compiler is only able to
     fold the builtin version (c.f. [2]).

For those two reasons, replace the assembly implementation by its
builtin counterpart.

[1] https://elixir.bootlin.com/linux/v6.0/source/include/asm-generic/bitops/builtin-__fls.h
[2] commit 146034fed6ee ("x86/asm/bitops: Use __builtin_ffs() to evaluate constant expressions")

CC: Borislav Petkov <bp@suse.de>
CC: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
---
 arch/x86/include/asm/bitops.h | 14 +-------------
 1 file changed, 1 insertion(+), 13 deletions(-)
diff mbox series

Patch

diff --git a/arch/x86/include/asm/bitops.h b/arch/x86/include/asm/bitops.h
index 2edf68475fec..a31453d7686d 100644
--- a/arch/x86/include/asm/bitops.h
+++ b/arch/x86/include/asm/bitops.h
@@ -285,19 +285,7 @@  static __always_inline unsigned long variable_ffz(unsigned long word)
 	 (unsigned long)__builtin_ctzl(~word) :	\
 	 variable_ffz(word))
 
-/*
- * __fls: find last set bit in word
- * @word: The word to search
- *
- * Undefined if no set bit exists, so code should check against 0 first.
- */
-static __always_inline unsigned long __fls(unsigned long word)
-{
-	asm("bsr %1,%0"
-	    : "=r" (word)
-	    : "rm" (word));
-	return word;
-}
+#include <asm-generic/bitops/builtin-__fls.h>
 
 #undef ADDR