arch/x86/include/asm/arch_hweight.h

   1 #ifndef _ASM_X86_HWEIGHT_H
   2 #define _ASM_X86_HWEIGHT_H
   3
   4 #include <asm/cpufeatures.h>
   5
   6 #ifdef CONFIG_64BIT
   7 /* popcnt %edi, %eax -- redundant REX prefix for alignment */
   8 #define POPCNT32 ".byte 0xf3,0x40,0x0f,0xb8,0xc7"
   9 /* popcnt %rdi, %rax */
  10 #define POPCNT64 ".byte 0xf3,0x48,0x0f,0xb8,0xc7"
  11 #define REG_IN "D"
  12 #define REG_OUT "a"
  13 #else
  14 /* popcnt %eax, %eax */
  15 #define POPCNT32 ".byte 0xf3,0x0f,0xb8,0xc0"
  16 #define REG_IN "a"
  17 #define REG_OUT "a"
  18 #endif
  19
  20 /*
  21  * __sw_hweightXX are called from within the alternatives below
  22  * and callee-clobbered registers need to be taken care of. See
  23  * ARCH_HWEIGHT_CFLAGS in <arch/x86/Kconfig> for the respective
  24  * compiler switches.
  25  */
  26 static __always_inline unsigned int __arch_hweight32(unsigned int w)
  27 {
  28         unsigned int res = 0;
  29
  30         asm (ALTERNATIVE("call __sw_hweight32", POPCNT32, X86_FEATURE_POPCNT)
  31                      : "="REG_OUT (res)
  32                      : REG_IN (w));
  33
  34         return res;
  35 }
  36
  37 static inline unsigned int __arch_hweight16(unsigned int w)
  38 {
  39         return __arch_hweight32(w & 0xffff);
  40 }
  41
  42 static inline unsigned int __arch_hweight8(unsigned int w)
  43 {
  44         return __arch_hweight32(w & 0xff);
  45 }
  46
  47 #ifdef CONFIG_X86_32
  48 static inline unsigned long __arch_hweight64(__u64 w)
  49 {
  50         return  __arch_hweight32((u32)w) +
  51                 __arch_hweight32((u32)(w >> 32));
  52 }
  53 #else
  54 static __always_inline unsigned long __arch_hweight64(__u64 w)
  55 {
  56         unsigned long res = 0;
  57
  58         asm (ALTERNATIVE("call __sw_hweight64", POPCNT64, X86_FEATURE_POPCNT)
  59                      : "="REG_OUT (res)
  60                      : REG_IN (w));
  61
  62         return res;
  63 }
  64 #endif /* CONFIG_X86_32 */
  65
  66 #endif