aarch64: Fix floating point retval recording

It seems that the address of 'float_retval' is loaded instead of its
return value, so it has to be dereferenced when it's passed to inline
assembly code.

Before:
  $ gcc -pg -g float.c
  $ uftrace -R float_add@retval/f32 -F float_add a.out
  0.100000
  # DURATION     TID     FUNCTION
     8.959 us [  4424] | float_add() = -593125376.000000;

After:
  $ gcc -pg -g float.c
  $ uftrace -R float_add@retval/f32 -F float_add a.out
  0.100000
  # DURATION     TID     FUNCTION
     1.303 us [  4653] | float_add() = 0.100000;

Signed-off-by: Honggyu Kim <[email protected]>

arch/aarch64/mcount-support.c

@@ -174,12 +174,12 @@ void mcount_arch_get_retval(struct mcount_arg_context *ctx,
  if (spec->size <= 4) {
  asm volatile ("ldr s0, %1\n"
  "str s0, %0\n" :
- "=m" (ctx->val.v) : "m" (float_retval));
+ "=m" (ctx->val.v) : "m" (*float_retval));
  }
  else {
  asm volatile ("ldr d0, %1\n"
  "str d0, %0\n" :
- "=m" (ctx->val.v) : "m" (float_retval));
+ "=m" (ctx->val.v) : "m" (*float_retval));
  }
  }
  else