Fix implementation of hwtime

[wierton]

In file matrix/matrix-android/matrix-sqlite-lint/src/lemon/os_common.h line 70, the implementation of hwtime seems to invoke assembly instruction rdtsc to get the cpu timestamp. From the specification of x86, rdtsc will output timestamp to register eax and edx, which can be written to a 64-bit variable by asm("rdtsc":"=A"(x)) directly, the mov %%edx, %%ecx is unused here, more than this, it may corrupt C variable under some context, for example:

__inline__ unsigned long long int hwtime(void){
  unsigned long long int x;
  __asm__("rdtsc\n\t"
          "mov %%edx, %%ecx\n\t"
          :"=A" (x));
  return x;
}

int __attribute__((noinline)) set_val(int v) { return v; }

int main() {
  int a = set_val(0);
  int b = set_val(1);
  int c = set_val(2);
  a = a + b; b = b + c; c = c + a;
  uint64_t d = hwtime();
  printf("%d, %d, %d, %lx\n", a, b, c, d);
  return 0;
}

Assume rdtsc now is 0x23232323, the output should be 1, 3, 3, 23232323, however the actual output on my PC (gcc-8 -O2) is 509053, 509055, 509055, 23232323, the first 3 numbers is random when execute. The reason is that the extra instruction mov %%edx, %%ecx overrides register ecx, this side effect is not realized by compiler (compiler will not parse inline assembly code to get exact semantics), unknown side effect results in incorrect schedule on register and cause C world breaks.

[tencent-adm]

All committers have signed the CLA.