Replacing the x86 popcntq Assembly Instruction

Symptom

Error: unknown mnemonic 'popcnt' -- 'popcnt [sp,8],x0'

Cause

popcnt is a bit 1 counting instruction of the x86 platform. The Kunpeng platform does not have a corresponding instruction, and a replacement algorithm needs to be used to implement the instruction.

Procedure

• Example code on the x86 platform:

  static inline uint64_t POPCNT_popcnt_u64(uint64_t a)
  { 
      uint64_t result; 
      __asm__("popcnt %1, %0" : "=r"(result) : "mr"(a) : "cc"); 
      return result; 
  }

• Example code on the Kunpeng platform:

  #include <arm_neon.h>
  static inline uint64_t POPCNT_popcnt_u64(uint64_t x)
  { 
      uint64_t count_result = 0; 
      uint64_t count[1]; 
      uint8x8_t input_val, count8x8_val; 
      uint16x4_t count16x4_val; 
      uint32x2_t count32x2_val; 
      uint64x1_t count64x1_val; 
      input_val = vld1_u8((unsigned char *) &x); 
      count8x8_val = vcnt_u8(input_val); 
      count16x4_val = vpaddl_u8(count8x8_val); 
      count32x2_val = vpaddl_u16(count16x4_val); 
      count64x1_val = vpaddl_u32(count32x2_val); 
      vst1_u64(count, count64x1_val); 
      count_result=count[0]; 
      return count_result; 
  }