lib/bch.c: Use __builtin_parity() when available
PATCH 1/1 lib/bch.c: Use __builtin_parity() when available
Compilers (GCC and clang) provide optimized __builtin_parity() function that returns the parity of X, i.e. the number of 1-bits in X modulo 2.
Use __builtin_parity() built-in function to optimize parity(). This improves generated code on x86_64 from:
movl %edi, %edx shrl %edx xorl %edi, %edx movl %edx, %eax shrl $2, %eax xorl %edx, %eax andl $286331153, %eax imull $286331153, %eax, %eax shrl $28, %eax andl $1, %eax
to an optimized:
movl %edi, %ecx shrl $16, %ecx xorl %edi, %ecx xorl %eax, %eax xorb %ch, %cl setnp %al
Please note SETNP instruction that exercises hardware parity calculation of x86 processors. When POPCNT instruction is available, the generated code gets optimized even further:
popcntl %edi, %eax
andl $1, %eax
Compile-tested only.
Signed-off-by Uros Bizjak <[email protected]>
Cc Andrew Morton <[email protected]>
--- lib/bch.c | 4 ++++ 1 file changed, 4 insertions(+)
lib/bch.c +4 -0
@@ -313,6 +313,9 @@ static inline int deg(unsigned int poly)static inline int parity (unsigned int x) { +#if __has_builtin (__builtin_parity) +return __builtin_parity (x); +#else /* * public domain code snippet, lifted from * http://www-graphics.stanford.edu/~seander/bithacks.html
@@ -321,6 +324,7 @@ static inline int parity(unsigned int x) x ^= x >>2 ; x = (x &0x11111111U ) *0x11111111U ;return (x >>28 ) &1 ; +#endif }/* Galois field basic operations: multiply, divide, inverse, etc. */
Signature · 1 line
2.42.0